%PDF-1.7
%
1 0 obj
<< /Metadata 3 0 R /Pages 4 0 R /Type /Catalog >>
endobj
2 0 obj
<< /Author (Jihyun Kim; Seong-Hun Jeong; Kyeongbo Kong; Suk-Ju Kang) /Producer (pikepdf 6.2.5) /Subject (IEEE Winter Conference on Applications of Computer Vision) /Title (An Unified Framework for Language Guided Image Completion) >>
endobj
3 0 obj
<< /Subtype /XML /Type /Metadata /Length 1169 >>
stream
An Unified Framework for Language Guided Image CompletionJihyun Kim Seong-Hun Jeong Kyeongbo Kong Suk-Ju KangIEEE Winter Conference on Applications of Computer Vision
endstream
endobj
4 0 obj
<< /Count 11 /Kids [ 5 0 R 6 0 R 7 0 R 8 0 R 9 0 R 10 0 R 11 0 R 12 0 R 13 0 R 14 0 R 15 0 R ] /Type /Pages >>
endobj
5 0 obj
<< /Contents 16 0 R /MediaBox [ 0 0 612 792 ] /Parent 4 0 R /Resources 17 0 R /Type /Page >>
endobj
6 0 obj
<< /Contents 18 0 R /MediaBox [ 0 0 612 792 ] /Parent 4 0 R /Resources 19 0 R /Type /Page >>
endobj
7 0 obj
<< /Contents 20 0 R /Group 21 0 R /MediaBox [ 0 0 612 792 ] /Parent 4 0 R /Resources 22 0 R /Type /Page >>
endobj
8 0 obj
<< /Contents 23 0 R /MediaBox [ 0 0 612 792 ] /Parent 4 0 R /Resources 24 0 R /Type /Page >>
endobj
9 0 obj
<< /Contents 25 0 R /MediaBox [ 0 0 612 792 ] /Parent 4 0 R /Resources 26 0 R /Type /Page >>
endobj
10 0 obj
<< /Contents 27 0 R /MediaBox [ 0 0 612 792 ] /Parent 4 0 R /Resources 28 0 R /Type /Page >>
endobj
11 0 obj
<< /Contents 29 0 R /MediaBox [ 0 0 612 792 ] /Parent 4 0 R /Resources 30 0 R /Type /Page >>
endobj
12 0 obj
<< /Contents 31 0 R /MediaBox [ 0 0 612 792 ] /Parent 4 0 R /Resources 32 0 R /Type /Page >>
endobj
13 0 obj
<< /Contents 33 0 R /MediaBox [ 0 0 612 792 ] /Parent 4 0 R /Resources 34 0 R /Type /Page >>
endobj
14 0 obj
<< /Contents 35 0 R /MediaBox [ 0 0 612 792 ] /Parent 4 0 R /Resources 36 0 R /Type /Page >>
endobj
15 0 obj
<< /Contents 37 0 R /MediaBox [ 0 0 612 792 ] /Parent 4 0 R /Resources 36 0 R /Type /Page >>
endobj
16 0 obj
<< /Length 11364 >>
stream
q
q
q
0 g 0 G
0 g 0 G
0 g 0 G
0 g 0 G
0 g 0 G
BT
/F75 14.3462 Tf 100.404 675.067 Td [(An)-250<556e69026564>-250(Framew)10(ork)-250(f)25(or)-250(Language)-250(Guided)-250(Image)-250(Completion)]TJ/F76 11.9552 Tf 21.603 -37.858 Td [(Jih)5(yun)-250(Kim)]TJ/F76 7.9701 Tf 56.069 2.577 Td [(*)]TJ/F36 7.9701 Tf 2.989 1.762 Td [(1)]TJ/F76 11.9552 Tf 4.732 -4.339 Td [(,)-250(Seong-Hun)-250(Jeong)]TJ/F76 7.9701 Tf 91.314 2.577 Td [(*)]TJ/F36 7.9701 Tf 4.483 1.762 Td [(2)]TJ/F76 11.9552 Tf 4.732 -4.339 Td [(,)-250(K)25(yeongbo)-250(K)35(ong)]TJ/F76 7.9701 Tf 84.619 2.577 Td [(*)]TJ/F36 7.9701 Tf 4.483 1.762 Td [(2)]TJ/F76 11.9552 Tf 4.732 -4.339 Td [(,)-250(and)-250(Suk-Ju)-250(Kang)]TJ/F36 7.9701 Tf 88.324 4.339 Td [(1)]TJ -305.539 -13.948 Td [(1)]TJ/F76 11.9552 Tf 4.733 -4.338 Td [(Sog)5(ang)-250(Uni)25(v)15(ersity)65(,)]TJ/F57 8.9664 Tf 93.991 0 Td [(f)]TJ/F78 8.9664 Tf 4.607 0 Td [(jhkim5950,)-600(sjkang)]TJ/F57 8.9664 Tf 91.457 0 Td [(g)]TJ/F78 8.9664 Tf 4.608 0 Td [(@sogang.ac.kr)]TJ/F36 7.9701 Tf -255.102 -9.61 Td [(2)]TJ/F76 11.9552 Tf 4.733 -4.338 Td [(Puk)15(yong)-250(National)-250(Uni)25(v)15(ersity)65(,)]TJ/F78 8.9664 Tf 144.68 0 Td [(tlqwkrk915@pukyong.ac.kr,)-600(kbkong@pknu.ac.kr)]TJ
0 g 0 G
/F75 11.9552 Tf -110.657 -41.046 Td [(Abstract)]TJ/F80 9.9626 Tf -83.928 -24.827 Td [(Ima)10(g)10(e)-276(completion)-276(is)-276(a)-276(r)37(esear)37(c)15(h)-276<02656c64>-276(whic)15(h)-276(aims)-276(to)-276(g)10(en-)]TJ -11.955 -11.955 Td [(er)15(ate)-261(vi)1(sual)-261(contents)-261(f)1(or)-261(unknown)-261(r)37(e)40(gions)-260(of)-261(an)-260(ima)10(g)10(e)15(.)-342(Im-)]TJ 0 -11.956 Td [(a)10(g)10(e)-284(outpainting)-285(and)-284(wide-r)15(ang)10(e)-284(ima)10(g)10(e)-285(bl)1(ending)10(,)-293(whic)15(h)-285(we)]TJ 0 -11.955 Td [(r)37(efer)-363(to)-362(as)-363(e)20(xtensive)-363(painting)10(,)-390(ar)37(e)-363(consider)37(ed)-363(c)15(hallenging)]TJ 0 -11.955 Td [(because)-277(compar)37(ed)-276(to)-277(the)-276(lar)37(g)10(e)-277(unknown)-277(r)37(e)40(gions)1(,)-284(r)37(elatively)]TJ 0 -11.955 Td [(less)-310(conte)20(xt)-310(is)-309(pr)45(o)10(vided.)-490(Some)-310(r)37(ecent)-309(studies)-310(have)-310(tried)-310(to)]TJ 0 -11.955 Td [(decr)37(ease)-371(the)-371(comple)20(xity)-372(of)-371(e)20(xtensive)-371(painting)-371(by)-372(g)10(ener)15(at-)]TJ 0 -11.955 Td [(ing)-334(ima)10(g)10(e)-334(hints)-334(for)-334(the)-333(missing)-334(r)37(e)40(gions.)-562(In)-334(this)-334(paper)111(,)-355(we)]TJ 0 -11.956 Td [(intr)45(oduce)-336(a)-336(no)10(vel)-335(modality)-336(of)-336(hints,)-357(the)-336(natur)15(al)-336(langua)10(g)10(e)15(.)]TJ 0 -11.955 Td [(Mor)37(eo)10(ver)111(,)-246(we)-245(pr)45(opose)-245(a)-245(Captioning-based)-246(Extensive)-245(P)80(aint-)]TJ 0 -11.955 Td [(ing)-327(\(CEP\))-328(module)10(,)-347(whic)15(h)-327(combines)-327(models)-328(for)-327(two)-328(dif)18(fer)20(-)]TJ 0 -11.955 Td [(ent)-397(multi-modal)-397(tasks:)-604(ima)10(g)10(e)-397(captioning)-397(and)-397(te)20(xt-guided)]TJ 0 -11.955 Td [(ima)10(g)10(e)-362(completion.)-645(In)-362(or)37(der)-361(to)-362(g)10(ener)15(ate)-362(appr)45(opriate)-362(cap-)]TJ 0 -11.955 Td [(tions)-242(for)-242(mask)10(ed)-243(ima)10(g)10(es,)-243(the)-243(ima)10(g)10(e)-242(captioning)-242(model)-242(is)-243(op-)]TJ 0 -11.956 Td [(timized)-223(using)-224(self-critical)-223(sequence)-224(tr)15(aining)-223(\(SCST\))-224(method)]TJ 0 -11.955 Td [(with)-338(r)15(andom)-339(masks.)-575(The)-338(big)10(g)10(est)-339<62656e650274>-338(of)-338(our)-339(methodol-)]TJ 0 -11.955 Td [(o)10(gy)-345(is)-344(the)-345(accessibility)-344(to)-345(well-designed)-344(ima)10(g)10(e)-345(captioning)]TJ 0 -11.955 Td [(and)-373(te)20(xt-guided)-372(ima)10(g)10(e)-373(manipulation)-372(models)-373(suc)15(h)-372(as)-373(OF)115(A)]TJ 0 -11.955 Td [(and)-389(GLIDE)-389(without)-390(the)-389(need)-389(for)-389(additional)-390(ar)37(c)15(hitectur)15(al)]TJ 0 -11.955 Td [(c)15(hang)10(es.)-683(In)-375(e)15(valuation,)-405(our)-375(model)-374(demonstr)15(ates)-375(r)37(emark-)]TJ 0 -11.956 Td [(able)-459(performance)-458(e)15(ven)-459(with)-458(complicated)-459(ima)10(g)10(e)-459(datasets)]TJ 0 -11.955 Td [(both)-250(quantitatively)-250(and)-250(qualitatively)55(.)]TJ/F75 11.9552 Tf 0 -38.615 Td [(1.)-250(Intr)18(oduction)]TJ/F76 9.9626 Tf 11.955 -19.846 Td [(Ev)15(eryone)-333(should)-332(be)-333(f)10(amiliar)-332(with)-333(Mona)-332(Lisa,)-353(the)-333(mas-)]TJ -11.955 -11.955 Td [(terpiece)-260(by)-261(Leonardo)-260(da)-261(V)60(inci,)-263(b)20(ut)-260(ha)20(v)15(e)-261(you)-260(e)25(v)15(er)-261(imagined)]TJ 0 -11.955 Td [(what)-393(kind)-392(of)-393(skirts)-393(and)-393(shoes)-392(Mona)-393(Lisa)-393(is)-392(wearing?)-739(If)]TJ 0 -11.955 Td [(so,)-242(you)-241(ha)20(v)15(e)-240(performed)-240(a)-241(type)-240(of)-240(image)-240(completion,)-243(kno)25(wn)]TJ 0 -11.955 Td [(more)-308<73706563690263616c6c79>-307(as)-308(image)-308(outpainting.)-483(Image)-308(completion)]TJ 0 -11.956 Td [(is)-333(a)-333(research)-333<02656c64>-332(which)-333(aims)-333(to)-333(generate)-333(plausible)-333(visual)]TJ 0 -11.955 Td [(contents)-410(for)-410(unkno)25(wn)-409(re)15(gions)-410(in)-410(gi)25(v)15(en)-410(images.)-789(This)-410(re-)]TJ 0 -11.955 Td [(search)-381(area)-380(includes)-381(not)-381(only)-381(con)40(v)15(entional)-381(ta)1(sks,)-414(such)-381(as)]TJ 0 -11.955 Td [(image)-251(inpainting)-251([1,)-251(2,)-251(3])-252(and)-251(outpainting)-251([4,)-251(5,)-251(6],)-252(b)20(ut)-251(also)]TJ 0 -11.955 Td [(a)-206(ne)25(wly)-205(suggested)-206(task,)-215(wi)1(de-range)-206(image)-206(blending)-205([7].)-296(Im-)]TJ 0 -11.955 Td [(age)-400(inpainting)-401(is)-400(an)-401(image)-400(restoration)-400(task)-401(which)-400<026c6c73>-401(in)]TJ 0 -11.956 Td [(missing/mask)10(ed)-221(re)15(gions)-220(of)-221(an)-221(input)-220(image.)-301(By)-220(contrast,)-227(im-)]TJ
0 g 0 G
ET
q
1 0 0 1 50.112 90.415 cm
[]0 d 0 J 0.398 w 0 0 m 94.499 0 l S
Q
BT
/F76 5.9776 Tf 60.971 82.492 Td [(*)]TJ/F76 7.9701 Tf 3.487 -1.492 Td [(equal)-250(contrib)20(ution)]TJ
0 g 0 G
0 g 0 G
0 g 0 G
/F76 9.9626 Tf 244.404 487.268 Td [(age)-201(outpainting)-202(is)-201(a)-202(task)-201(which)-202(aims)-201(to)-201(e)15(xtend)-202(a)-201(gi)25(v)15(en)-202(image)]TJ 0 -11.955 Td [(be)15(yond)-360(its)-359(original)-360(boundaries.)-638(Lastly)65(,)-387(wide-range)-360(image)]TJ 0 -11.956 Td [(blending)-349(aims)-348(to)-349(generate)-349(an)-348(intermediate)-349(image)-349(between)]TJ 0 -11.955 Td [(tw)10(o)-320(dif)25(ferent)-320(images)-320(so)-319(that)-320(the)15(y)-320(form)-320(a)-320(single)-320(panoramic)]TJ 0 -11.955 Td [(image.)-505(Image)-314(outpainting)-315(and)-315(wide-range)-315(image)-315(blending)]TJ 0 -11.955 Td [(ha)20(v)15(e)-349(been)-349(pro)15(v)15(en)-349(to)-349(be)-349(more)-350(dif)26<0263756c74>-350(compared)-349(to)-349(image)]TJ 0 -11.955 Td [(inpainting)-369(because)-368(the)15(y)-369(create)-369(no)15(v)15(el)-369(contents)-368(for)-369(a)-369(lar)18(ger)]TJ 0 -11.955 Td [(area)-278(and)-278(ha)20(v)15(e)-279(less)-278(conte)15(xtual)-278(information)-278(to)-279(t)1(ak)10(e)-279(reference)]TJ 0 -11.956 Td [(from.)-606(As)-349(such,)-373(we)-349(ha)20(v)15(e)-348(decided)-349(that)-348(tackling)-349(image)-349(out-)]TJ 0 -11.955 Td [(painting)-364(and)-363(wide-range)-364(image)-364(blending)-364(w)10(ould)-363(be)-364(an)-364(e)15(x-)]TJ 0 -11.955 Td [(cellent)-334(w)10(ay)-334(to)-335(demonstrate)-334(the)-334(competenc)15(y)-334(and)-335(ef)25<026369656e63>15(y)]TJ 0 -11.955 Td [(of)-342(our)-342(method)-341(\(Fig.)-586(1\).)-585(F)15(or)-342(con)40(v)15(enience,)-364(we)-342(refer)-342(to)-342(the)]TJ 0 -11.955 Td [(tw)10(o)-250(challenging)-250(tasks)-250(as)]TJ/F80 9.9626 Tf 99.516 0 Td [(e)20(xtensive)-250(painting)]TJ/F76 9.9626 Tf 72.019 0 Td [(.)]TJ -159.58 -16.486 Td [(Recently)65(,)-280(there)-273(ha)20(v)15(e)-274(been)-274(ef)25(forts)-273(to)-274(solv)15(e)-274(the)-273(lack)-274(of)-274(in-)]TJ -11.955 -11.955 Td [(formation)-284(problem)-284(by)-284(generating)-284(image)-284(hints)-284(for)-284(the)-284(miss-)]TJ 0 -11.955 Td [(ing)-296(re)15(gions.)-448(One)-295(of)-296(the)-296(techniques)-296(that)-296(has)-296(been)-296(instituted)]TJ 0 -11.956 Td [(to)-374(resolv)15(e)-375(thi)1(s)-375(issue)-374(is)-374(Bidirectional)-374(Rearrangement)-375(\(BR\))]TJ 0 -11.955 Td [([8],)-314(which)-301(utilize)1(s)-301(the)-301(opposite)-301(parts)-301(of)-301(the)-301(input)-301(image)-301(as)]TJ 0 -11.955 Td [(hints)-308(by)-309(switching)-308(the)-309(left)-308(and)-309(rights)-308(sides,)-323(and)-309<026c6c73>-308(in)-309(the)]TJ 0 -11.955 Td [(g)5(ap)-403(between)-403(the)-402(tw)10(o)-403(split)-403(images.)-768(In)-403(addition,)-441(the)-403(Mir)20(-)]TJ 0 -11.955 Td [(rored)-327(Rearrangement)-327(\(MR\))-327([9])-327(predicts)-327(missing)-327(re)15(gion)-327(by)]TJ 0 -11.955 Td [(using)-234(the)-234(mirror)-233<036970706564>-234(input)-234(image)-234(as)-234(hints.)-304(While)-234(the)-234(im-)]TJ 0 -11.956 Td [(age)-311(hints)-310(made)-311(by)-310(the)-311(BR)-311(and)-310(the)-311(MR)-310(methods)-311(ha)20(v)15(e)-311(only)]TJ 0 -11.955 Td [(been)-201(applicable)-201(to)-201(symmetrical)-201(images)-201(due)-201(to)-201(structural)-201(rea-)]TJ 0 -11.955 Td [(sons,)-229(a)-223(more)-223(recent)-223(study)65(,)-229(Image-Adapti)25(v)15(e)-223(Hint)-223(\(IAH\))-224([10],)]TJ 0 -11.955 Td [(e)15(xpands)-358(upon)-359(the)-358(hint-based)-359(method)-358(to)-358(asymmetrical)-359(im-)]TJ 0 -11.955 Td [(ages.)-344(IAH)-262(generates)-261(image)-262(hints)-261(via)-262(V)60(ision)-261(T)35(ransformer)-262(in)]TJ 0 -11.955 Td [(an)-348(image-adapti)25(v)15(e)-348(w)10(ay)65(.)-605(Although)-348(v)25(arious)-348(hint-generation)]TJ 0 -11.956 Td [(methods)-311(ha)20(v)15(e)-310(been)-311(implemented,)-326(the)-311(output)-310(hints)-311(are)-311(lim-)]TJ 0 -11.955 Td [(ited)-250(to)-250(the)-250(image)-250(format.)]TJ 11.955 -16.485 Td [(In)-272(this)-272(paper)40(,)-278(we)-272(introduce)-272(a)-272(no)15(v)15(el)-272(modality)-272(of)-272(hints)-272(for)]TJ -11.955 -11.956 Td [(e)15(xtensi)25(v)15(e)-257(painting)-257(tasks,)-258(the)-257(natural)-257(language.)-330(As)-257(sho)25(wn)-257(in)]TJ 0 -11.955 Td [(Fig.)-428(2,)-300(to)-289(incorporate)-290(language)-289(hints)-290(to)-289(e)15(xtensi)25(v)15(e)-290(painting,)]TJ 0 -11.955 Td [(our)-325<756e69026564>-325(frame)25(w)10(ork)-325(named)-326(Captioning-based)-325(Extensi)25(v)15(e)]TJ 0 -11.955 Td [(P)15(ainting)-312(\(CEP\))-311(module)-312(consists)-312(of)-311(tw)10(o)-312(multi-modal)-312(tasks:)]TJ 0 -11.955 Td [(an)-286(image)-286(captioning)-286(task)-286(and)-286(a)-286(te)15(xt-guided)-286(image)-286(manipu-)]TJ 0 -11.955 Td [(lation)-283(task.)-408(Firstly)65(,)-291(the)-283(image)-283(captioning)-283(module)-283(generates)]TJ 0 -11.956 Td [(language)-311(hints)-311(by)-311(capturing)-311(semantic)-311(and)-311(te)15(xtual)-311(informa-)]TJ 0 -11.955 Td [(tion)-218(of)-218(the)-217(gi)25(v)15(en)-218(images.)-299(Secondly)65(,)-225(with)-217(the)-218(guidance)-218(of)-218(the)]TJ 0 -11.955 Td [(language)-237(hints,)-240(the)-237(te)15(xt-guided)-237(image)-237(manipulation)-237(module)]TJ 0 -11.955 Td [(generates)-231(visual)-230(contents)-231(for)-231(mask)10(ed)-231(re)15(gions)-230(that)-231(are)-231(in)-231(line)]TJ
0 g 0 G
0 g 0 G
ET
Q
Q
q
1 0 0 1 0 0 cm
/RXQNdy4_Gr9yUkuYqjE_FQ Do
Q
Q
q
1 0 0 1 0 0 cm
/Qzccfj9BXudNHzIg6T-X5Q Do
Q
endstream
endobj
17 0 obj
<< /Font << /F36 38 0 R /F57 39 0 R /F75 40 0 R /F76 41 0 R /F78 42 0 R /F80 43 0 R >> /ProcSet [ /PDF /Text ] /XObject << /Qzccfj9BXudNHzIg6T-X5Q 44 0 R /RXQNdy4_Gr9yUkuYqjE_FQ 45 0 R >> >>
endobj
18 0 obj
<< /Length 9117 >>
stream
q
q
0 g 0 G
0 g 0 G
0 g 0 G
1 0 0 1 50.112 569.952 cm
q
.51581 0 0 .51581 0 0 cm
q
959.6778 0 0 290.8993 0 0 cm
/Im1 Do
Q
Q
1 0 0 1 -50.112 -569.952 cm
BT
/F76 8.9664 Tf 50.112 547.038 Td [(Figure)-278(1.)]TJ/F75 8.9664 Tf 37.357 0 Td [(Illustration)-250(of)-250(image)-250(outpainting)-250(and)-250(wide-range)-250(image)-250(blending)15(.)]TJ/F76 8.9664 Tf 251.94 0 Td [(Image)-278(outpainting)-277(is)-278(a)-278(task)-277(which)-278(aims)-278(to)-277(e)15(xtend)-278(a)-278(gi)25(v)15(en)]TJ -289.297 -10.959 Td [(image)-224(be)15(yond)-224(its)-225(original)-224(boundaries.)-301(W)40(ide-range)-224(image)-224(blending)-225(aims)-224(to)-224(generate)-224(an)-224(intermediate)-224(image)-225(between)-224(tw)10(o)-224(dif)25(ferent)-224(images)-224(so)]TJ 0 -10.959 Td [(that)-214(t)1(he)15(y)-214(form)-214(a)-213(single)-214(panoramic)-213(image.)-298(Compared)-213(to)-214(the)-213(state-of-the-art)-214(algorithm)-213(of)-214(each)-213(task,)-221(our)-214(method)-213(has)-214(outstanding)-213(performance.)]TJ 0 -10.959 Td [(Red)-250(box)15(es)-250(indicate)-250(generated)-250(re)15(gions.)]TJ
0 g 0 G
/F76 9.9626 Tf 0 -31.829 Td [(with)-394(the)-394(content)-394(of)-395(the)-394(unmask)10(ed)-394(re)15(gions.)-742(F)15(or)-395(each)-394(im-)]TJ 0 -11.955 Td [(age)-233(outpainting)-232(task,)-237(hint)-232(generation)-233(and)-233(hint-guided)-233(image)]TJ 0 -11.955 Td [(editing)-332(are)-332(each)-331(p)-1(e)1(rformed)-332(once.)-556(On)-332(the)-332(contrary)65(,)-352(for)-332(the)]TJ 0 -11.955 Td [(wide-range)-311(i)1(mage)-311(blending)-311(task,)-325(image)-311(outpainting)-310(on)-311(the)]TJ 0 -11.955 Td [(tw)10(o)-274(input)-274(images)-274(is)-275(repeated)-274(until)-274(the)-274(g)5(ap)-274(between)-275(the)-274(tw)10(o)]TJ 0 -11.956 Td [(images)-345(is)-345(completely)-346<026c6c65642e>-596(In)-345(the)-345(end,)-369(to)-345(ensure)-346(smooth)]TJ 0 -11.955 Td [(transition,)-240(our)-238(frame)25(w)10(ork)-239(perfor)1(ms)-239(image)-238(inpainting)-238(on)-238(the)]TJ 0 -11.955 Td [(v)15(ery)-250(middle)-250(part)-250(of)-250(the)-250(generated)-250(intermediate)-250(image.)]TJ 11.955 -15.361 Td [(The)-384(most)-385(important)-384(part)-384(of)-385(our)-384(CEP)-384(module)-384(is)-385(image)]TJ -11.955 -11.955 Td [(captioning.)-410(No)-283(matter)-283(e)15(xcellent)-283(the)-283(te)15(xt-guided)-284(image)-283(ma-)]TJ 0 -11.955 Td [(nipulation)-230(models)-231(are,)-234(the)-231<026e616c>-230(results)-230(of)-231(our)-230<756e69026564>-231(frame-)]TJ 0 -11.956 Td [(w)10(ork)-242(hea)20(vily)-242(depend)-242(on)-243(the)-242(captions.)-307(Nonetheless,)-244(e)15(xisting)]TJ 0 -11.955 Td [(captioning)-309(models)-309(are)-310(incapable)-309(of)-309(generating)-309(proper)-310(cap-)]TJ 0 -11.955 Td [(tions)-235(for)-234(the)-235(mask)10(ed)-235(images)-235(because)-234(the)15(y)-235(ha)20(v)15(e)-235(been)-235(trained)]TJ 0 -11.955 Td [(only)-447(on)-447(complete)-447(images.)-901(F)15(or)-447(instance,)-497(during)-447(our)-447(e)15(x-)]TJ 0 -11.955 Td [(periment,)-388(the)-361(captioning)-360(model)-361(repeatedly)-360(recognized)-361(the)]TJ 0 -11.955 Td [(mask)10(ed)-390(image)-391(as)-390(a)-391(combination)-390(of)-391(tw)10(o)-390(dif)25(ferent)-391(images,)]TJ 0 -11.956 Td [(and)-321(generated)-320(inappropriate)-321<7072650278>15(es)-321(such)-320(as)-321(`tw)10(o)-321(pictures)]TJ 0 -11.955 Td [(of)-55(',)-436(resulting)-398(in)-399(generation)-398(of)-399(semantically)-398(a)15(wkw)10(ard)-399(im-)]TJ 0 -11.955 Td [(ages.)-426(T)80(o)-288(solv)15(e)-289(this)-289(problem,)-298(we)-288(added)-289(to)-289(our)-288(CEP)-289(module)]TJ 0 -11.955 Td [(an)-323(optimizing)-324(process)-323(using)-323(self-critical)-323(sequence)-324(training)]TJ 0 -11.955 Td [(\(SCST\))-413(method)-413(with)-413(randomly)-413(mask)10(ed)-413(images)-414(and)-413(their)]TJ 0 -11.955 Td [(matching)-360(captions.)-640(As)-360(a)-360(res)1(ult,)-388(our)-360(optimized)-360(captioning)]TJ 0 -11.956 Td [(model)-356(is)-356(able)-357(to)-356(predict)-356(an)-356(appropriate)-357(ca)1(p)-1(t)1(ion)-357(re)15(g)5(ardless)]TJ 0 -11.955 Td [(of)-297(the)-297(mask)-297(shapes.)-450(Most)-297(importantly)65(,)-309(it)-297(can)-297(generate)-297(lan-)]TJ 0 -11.955 Td [(guage)-264(hints)-263(for)-264(v)25(arious)-264(image)-263(completion)-264(tasks,)-267(which)-264(w)10(as)]TJ 0 -11.955 Td [(ne)25(v)15(er)-220(possible)-220(for)-220(prior)-221(hint-based)-220(methods)-220([8,)-220(9,)-220(10])-221(due)-220(to)]TJ 0 -11.955 Td [(their)-250(dependenc)15(y)-250(on)-250(image)-250(structures.)]TJ 11.955 -15.361 Td [(Another)-241<62656e650274>-242(of)-241(our)-242(module)-241(is)-242(the)-241(accessibility)-241(to)-242(v)25(ar)20(-)]TJ -11.955 -11.956 Td [(ious)-354(pretrained)-353(models.)-622(Since)-353(both)-354(image)-354(captioning)-354(and)]TJ 0 -11.955 Td [(te)15(xt-guided)-468(image)-468(manipulation)-468(ha)20(v)15(e)-468(been)-468(popular)-468(e)25(v)15(en)]TJ 0 -11.955 Td [(among)-517(man)15(y)-518(multi-modal)-517(tasks,)-585(v)25(arious)-517(algorithms)-518(are)]TJ 0 -11.955 Td [(a)20(v)25(ailable.)-301(F)15(or)-222(e)15(xample,)-228(popular)-222(algorithms)-223(such)-222(as)-222(One)-223(F)15(or)]TJ 0 -11.955 Td [(All)-396(\(OF)74(A\))-396([11])-396(and)-396(ClipCap)-396([12])-396(are)-396(for)-396(the)-396(image)-396(cap-)]TJ
0 g 0 G
0 g 0 G
258.75 401.332 Td [(tioning)-401(task,)-438(while)-401(GLIDE)-401([13])-401(and)-401(a)-401(te)15(xt-dri)25(v)15(e)-401(blended)]TJ 0 -11.955 Td [(dif)25(fusion)-370([14])-370(are)-370(for)-370(the)-371(te)15(xt-guided)-370(image)-370(manipulation)]TJ 0 -11.955 Td [(task.)-380(All)-274(of)-273(the)-274(algorithms)-273(mentioned)-274(abo)15(v)15(e)-273(ha)20(v)15(e)-274(achie)25(v)15(ed)]TJ 0 -11.955 Td [(a)-252(remarkable)-252(le)25(v)15(el)-252(of)-252(performance.)-315(In)-252(particular)40(,)-253(trained)-252(on)]TJ 0 -11.956 Td [(hundreds)-269(of)-270(millions)-269(of)-269(data,)-274(lar)18(ge-scale)-269(te)15(xt-guided)-270(image)]TJ 0 -11.955 Td [(synthesis)-342(models)-342(are)-342(capable)-342(of)-342(generating)-342(photo-realistic)]TJ 0 -11.955 Td [(images)-310(across)-309(most)-310(of)-309(the)-310(image)-310(domains)-309([15].)-489(By)-310(taking)]TJ 0 -11.955 Td [(adv)25(antage)-256(of)-255(these)-256(multi-modal)-256(models,)-257(our)-255(module)-256(is)-256(able)]TJ 0 -11.955 Td [(to)-325(generate)-325(semantically)-324(plausible)-325(and)-325(realistic)-325(images)-325(for)]TJ 0 -11.955 Td [(image)-300(outpainting,)-313(as)-300(well)-300(as)-300(for)-300(wide-range)-300(image)-300(blend-)]TJ 0 -11.956 Td [(ing.)]TJ 11.955 -12.326 Td [(In)-250(essence,)-250(our)-250(main)-250(contrib)20(utions)-250(include:)]TJ
0 g 0 G
-0.498 -23.4 Td [<95>]TJ
0 g 0 G
[-500(W)80(e)-201(introduce)-202(natural)-201(language)-202(hints)-201(for)-201(image)-202(comple-)]TJ 8.468 -11.955 Td [(tion)-208(tasks,)-217<73706563690263616c6c79>-208(for)-208(image)-208(outpainting)-209(and)-208(wide-)]TJ 0 -11.956 Td [(range)-235(image)-234(blending.)-305(W)80(e)-235(refer)-235(to)-234(the)-235(tw)10(o)-235(challenging)]TJ 0 -11.955 Td [(tasks)-352(as)]TJ/F80 9.9626 Tf 35.23 0 Td [(e)20(xtensive)-352(painting)]TJ/F76 9.9626 Tf 76.534 0 Td [(in)-352(reference)-351(to)-352(their)-351(lar)18(ge)]TJ -111.764 -11.955 Td [(unkno)25(wn)-266(re)15(gions.)-359(Although)-267(there)-266(ha)20(v)15(e)-266(been)-266(ef)25(forts)-267(to)]TJ 0 -11.955 Td [(utilize)-253(image)-252(hints)-253(for)-253(image)-252(outpainting)-253(tasks,)-253(we)-253(are)]TJ 0 -11.955 Td [(the)-250<02727374>-250(to)-250(generate)-250(hints)-250(in)-250(a)-250(language)-250(format.)]TJ
0 g 0 G
-8.468 -21.408 Td [<95>]TJ
0 g 0 G
[-500(W)80(e)-220(optimized)-220(an)-220(image)-220(captioning)-220(model)-220(with)-220(a)-220(lar)18(ge-)]TJ 8.468 -11.956 Td [(scale)-352(caption)-352(dataset,)-378(using)-352(random)-352(masks.)-616(As)-353(a)-352(re-)]TJ 0 -11.955 Td [(sult,)-355(our)-333(image)-334(captioning)-333(model)-334(generates)-334(appropri-)]TJ 0 -11.955 Td [(ate)-245(language)-245(hints)-244(for)-245(v)25(arious)-245(image)-245(completion)-245(tasks,)]TJ 0 -11.955 Td [(re)15(g)5(ardless)-250(of)-250(the)-250(mask)-250(shapes.)]TJ
0 g 0 G
-8.468 -21.408 Td [<95>]TJ
0 g 0 G
[-500(Our)-333(frame)25(w)10(ork)-334(can)-333(utilize)-334(v)25(arious)-333(algorithms)-333(for)-334(the)]TJ 8.468 -11.955 Td [(image)-319(captioning)-319(task)-318(and)-319(the)-319(te)15(xt-guided)-319(image)-319(ma-)]TJ 0 -11.956 Td [(nipulation)-489(task,)-548(and)-489(links)-489(them)-488(together)-489(for)-489(e)15(xten-)]TJ 0 -11.955 Td [(si)25(v)15(e)-306(painting.)-477(Thanks)-305(to)-306(the)-306(recent)-305(progress)-306(in)-306(multi-)]TJ 0 -11.955 Td [(modal)-320(tasks,)-338(especially)-320(in)-320(the)-320(te)15(xt-guided)-321(image)-320(syn-)]TJ 0 -11.955 Td [(thesis)-424(task,)-466(o)-1(ur)-423(module)-424(achie)25(v)15(es)-423(remarkable)-424(perfor)20(-)]TJ 0 -11.955 Td [(mance)-366(and)-365(high)-366(le)25(v)15(el)-365(of)-366(generalization)-366(in)-365(image)-366(do-)]TJ 0 -11.955 Td [(mains.)]TJ
0 g 0 G
0 g 0 G
ET
Q
Q
q
1 0 0 1 0 0 cm
/Ffqmz3rv7UPkdhUCrllcGg Do
Q
endstream
endobj
19 0 obj
<< /Font << /F75 40 0 R /F76 41 0 R /F80 43 0 R >> /ProcSet [ /PDF /Text /ImageC ] /XObject << /Ffqmz3rv7UPkdhUCrllcGg 46 0 R /Im1 47 0 R >> >>
endobj
20 0 obj
<< /Length 8532 >>
stream
q
q
0 g 0 G
0 g 0 G
0 g 0 G
1 0 0 1 61.362 511.626 cm
q
1.42758 0 0 1.42758 0 0 cm
q
330.9818 0 0 145.9636 0 0 cm
/Im2 Do
Q
Q
1 0 0 1 -61.362 -511.626 cm
BT
/F76 8.9664 Tf 50.112 495.685 Td [(Figure)-223(2.)]TJ/F75 8.9664 Tf 34.133 0 Td [(Ov)10(erall)-250(pr)18(ocess)-250(of)-250(extensi)10(v)10(e)-250(painting)-250(with)-250(the)-250(pr)18(oposed)-250(CEP)-250(module.)]TJ/F97 8.9664 Tf 265.913 0 Td [(Left)]TJ/F76 8.9664 Tf 14.937 0 Td [(:)-296(Image)-223(outpainting)-223(is)-223(performed)-223(through)-223(our)-223(CEP)]TJ -314.983 -10.959 Td [(module)-280(which)-280(consists)-280(of)-281(image)-280(captioning)-280(and)-280(te)15(xt-guided)-280(image)-280(manipulation)-281(netw)10(orks.)-400(A)-280(mask)10(ed)-280(image)-281(is)-280(passed)-280(through)-280(the)-280(image)]TJ 0 -10.959 Td [(captioning)-250(netw)10(ork)-250(to)-250(generate)-250(the)-250(language)-250(hint.)-310(Then,)-250(with)-250(the)-250(language)-250(hint,)-250(the)-250(missing)-250(re)15(gions)-250(are)-250(predicted)-250(by)-250(an)-250(te)15(xt-guided)-250(image)]TJ 0 -10.958 Td [(manipulation)-299(netw)10(ork.)]TJ/F97 8.9664 Tf 85.646 0 Td [(Right)]TJ/F76 8.9664 Tf 20.434 0 Td [(:)-409(W)40(ide-range)-299(Image)-299(blending)-299(consists)-299(of)-300(three)-299(stages.)-458(First)-299(tw)10(o)-299(stages)-299(repeat)-300(image)-299(outpainting)-299(for)]TJ/F67 8.9664 Tf 365.678 0 Td [(N)]TJ/F76 8.9664 Tf 8.335 0 Td [(,)]TJ/F67 8.9664 Tf 5.036 0 Td [(M)]TJ/F76 8.9664 Tf -485.129 -10.959 Td [(times)-295(on)-295(dif)25(ferent)-295(images)-295(with)-295(the)-295(re)25(v)15(erse)-296(direc)1(tion.)-446(F)15(or)-295(a)-295(natural)-295(connection,)-306(the)-296(pr)1(edicted)-296(output)-295(from)-295(the)-295(pre)25(vious)-295(step)-295(is)-295(used)-295(as)-295(the)]TJ 0 -10.959 Td [(input)-342(for)-343(the)-342(ne)15(xt)-342(step.)-587(Then,)-366(to)-342(blend)-342(the)-343(resulted)-342(images,)-365(in)-343(the)-342(image)-342(blending)-343(stage,)-365(we)-342(mask)-343(the)-342(disconnected)-342(re)15(gion)-343(and)-342(pass)-342(it)]TJ 0 -10.959 Td [(through)-250(our)-250(CEP)-250(module)-250(once.)]TJ
0 g 0 G
0 g 0 G
0 g 0 G
/F75 11.9552 Tf 0 -31.881 Td [(2.)-250(Related)-250(W)75(ork)]TJ/F75 10.9589 Tf 0 -19.467 Td [(2.1.)-250(Image)-250(Completion)]TJ/F76 9.9626 Tf 11.955 -18.47 Td [(The)-378(tw)10(o)-378(major)-379(tasks)-378(of)-378(image)-378(completion)-379(i)1(nclude)-379(im-)]TJ -11.955 -11.956 Td [(age)-329(inpainting)-328(and)-329(image)-328(outpainting.)-546(Classic)-329(approaches)]TJ 0 -11.955 Td [(for)-241(both)-241(tasks)-241(were)-242(mostly)-241(patch-based)-241([16,)-241(17,)-241(18,)-241(19,)-242(20],)]TJ 0 -11.955 Td [(which)-388<026e64>-387(patches)-388(from)-388(kno)25(wn)-388(re)15(gions)-387(for)-388(each)-388(mask)10(ed)]TJ 0 -11.955 Td [(re)15(gion.)-704(Ho)25(we)25(v)15(er)40(,)-414(these)-381(methods)-381(often)-381(f)10(ailed)-381(to)-382(learn)-381(se-)]TJ 0 -11.955 Td [(mantic)-301(structures)-301(and)-301(had)-301(limited)-301(abilities)-301(to)-302(generate)-301(ne)25(w)]TJ 0 -11.955 Td [(contents.)-359(On)-266(the)-267(contrary)65(,)-270(learning-based)-267(methods)-266([1,)-266(2,)-267(3])]TJ 0 -11.956 Td [(were)-384(relati)25(v)15(ely)-384(better)-383(at)-384(capturing)-384(semantics.)-711(As)-384(for)-384(im-)]TJ 0 -11.955 Td [(age)-231(outpainting,)-234(due)-230(to)-231(the)-231(lack)-230(of)-231(adjacent)-230(information)-231(and)]TJ 0 -11.955 Td [(lar)18(ger)-243(unkno)25(wn)-243(area,)-244(there)-243(ha)20(v)15(e)-243(bee)1(n)-243(ef)25(forts)-243(to)-243(generate)-243(vi-)]TJ 0 -11.955 Td [(sual)-319(hints)-319(for)-319(the)-319(missing)-320(re)15(gions)-319([8,)-319(9,)-319(10].)-517(Recently)65(,)-337([7])]TJ 0 -11.955 Td [(proposed)-314(a)-314(ne)25(w)-315(image)-314(completion)-314(task,)-330(wide-range)-315(image)]TJ 0 -11.956 Td [(blending,)-413(which)-380(aims)-380(to)-381(mer)18(ge)-380(tw)10(o)-380(dif)25(ferent)-380(images)-381(into)]TJ 0 -11.955 Td [(one)-292(by)-293(generating)-292(appropriate)-293(contents)-292(in)-293(between)-292(the)-293(tw)10(o)]TJ 0 -11.955 Td [(images.)-536(F)15(or)-325(this)-325(task,)-344([7])-325(presents)-325(an)-326(encoder)20(-de)1(coder)-326(ar)20(-)]TJ 0 -11.955 Td [(chitecture,)-231(which)-225(sequentially)-226(predicts)-226(the)-226(feature)-226(represen-)]TJ 0 -11.955 Td [(tations)-250(for)-250(the)-250(intermediate)-250(re)15(gion.)]TJ/F75 10.9589 Tf 0 -21.54 Td [(2.2.)-250(Image)-250(captioning)]TJ/F76 9.9626 Tf 11.955 -18.47 Td [(Modern)-393(image)-393(captioning)-393(models)-393(typically)-394(emplo)10(y)-393(an)]TJ -11.955 -11.956 Td [(encoder)20(-decoder)-276(architecture)-276([21,)-276(22,)-276(23],)-282(in)-276(which)-276(the)-276(en-)]TJ 0 -11.955 Td [(coder)-397(e)15(xtracts)-396(visual)-397(features)-396(from)-397(an)-396(image)-397(and)-396(the)-397(de-)]TJ 0 -11.955 Td [(coder)-268(generates)-268(a)-269(sequence)-268(of)-268(w)10(ords)-268(from)-268(the)-269(e)15(xtracted)-268(vi-)]TJ 0 -11.955 Td [(sual)-228(features.)-303(T)80(o)-228(train)-228(an)-228(image)-228(captioning)-228(model,)-233(cross)-228(en-)]TJ
0 g 0 G
0 g 0 G
258.75 317.051 Td [(trop)10(y)-282(loss)-281(follo)25(wed)-282(by)-281(reinforcement)-282(learning)-282([24])-281(is)-282(com-)]TJ 0 -11.955 Td [(monly)-407(used.)-782(Such)-407(training)-408(method)-407(enables)-407(the)-407(usage)-408(of)]TJ 0 -11.955 Td [(non-dif)25(ferentiable)-449(capti)1(o)-1(n)-448(metrics)-449(as)-448(optimization)-449(objec-)]TJ 0 -11.955 Td [(ti)25(v)15(es.)-359(Recently)65(,)-270(there)-266(were)-266(ef)25(forts)-266(to)-267(adopt)-266(vision-language)]TJ 0 -11.955 Td [(pretraining)-282(on)-282(lar)18(ge)-281(image-te)15(xt)-282(corpus)-282(to)-282(image)-282(captioning)]TJ 0 -11.956 Td [(tasks)-250([25,)-250(26].)]TJ/F75 10.9589 Tf 0 -18.55 Td [(2.3.)-250(T)92(extguided)-250(Image)-250(Manipulation)]TJ/F76 9.9626 Tf 11.955 -17.933 Td [(Initially)-217(te)15(xt-guided)-217(image)-217(manipulation)-218(models)-217(focused)]TJ -11.955 -11.955 Td [(on)-338(editing)-339(an)-338(image)-338(according)-338(to)-339(a)-338(gi)25(v)15(en)-338(te)15(xt)-338(prompt)-339(and)]TJ 0 -11.955 Td [(the)-419(model)-419(decided)-419(which)-419(part)-419(of)-419(the)-419(image)-419(to)-419(edit)-419([27].)]TJ 0 -11.955 Td [([28])-446(introduces)-446(a)-446(te)15(xt-guided)-446(image)-446(manipulation)-446(model)]TJ 0 -11.956 Td [(which)-349(generates)-348(contents)-349(only)-348(for)-349(desired)-349(re)15(gions)-348(using)-349(a)]TJ 0 -11.955 Td [(dual)-361(attention)-362(mechanism.)-644(Recent)-362(w)10(orks)-361(further)-362(impro)15(v)15(e)]TJ 0 -11.955 Td [(the)-242(performance)-243(of)-242(te)15(xt-guided)-243(image)-242(manipulation)-243(models)]TJ 0 -11.955 Td [(by)-250(training)-250(them)-250(on)-250(lar)18(ge)-250(scale)-250(datasets)-250([13,)-250(14].)]TJ/F75 11.9552 Tf 0 -22.536 Td [(3.)-250(Pr)18(oposed)-250(Method)]TJ/F76 9.9626 Tf 11.955 -18.929 Td [(In)-386(this)-386(s)1(ection,)-420(we)-386(pro)15(vide)-386(an)-385(o)15(v)15(ervie)25(w)-386(of)-386(our)-386(frame-)]TJ -11.955 -11.955 Td [(w)10(ork,)-381(which)-355(is)-355(sho)25(wn)-355(in)-355(Fig.)-625(2.)-625(Our)-355(frame)25(w)10(ork,)-381(named)]TJ 0 -11.955 Td [(Captioning-based)-323(Ext)1(ensi)25(v)15(e)-323(P)15(ainting)-323(\(CEP\))-322(module,)-341(com-)]TJ 0 -11.955 Td [(prises)-366(tw)10(o)-366(types)-366(of)-365(netw)10(orks)-366(1\))-366(an)-366(image)-366(captioning)-366(net-)]TJ 0 -11.956 Td [(w)10(ork)]TJ/F23 9.9626 Tf 24.26 0 Td [(G)]TJ/F22 6.9738 Tf 7.833 -1.494 Td [(C)-70(AP)]TJ/F76 9.9626 Tf 22.723 1.494 Td [(for)-390(language)-390(hints)-390(generation)-390(and)-390(2\))-390(a)-391(te)15(xt-)]TJ -54.816 -11.955 Td [(guided)-207(image)-207(manipulation)-206(netw)10(ork,)]TJ/F23 9.9626 Tf 147.416 0 Td [(G)]TJ/F22 6.9738 Tf 7.833 -1.494 Td [(I)-78(M)]TJ/F76 9.9626 Tf 12.916 1.494 Td [(.)-296(Basically)65(,)-215(when)]TJ -168.165 -11.955 Td [(a)-283(mask)10(ed)-283(image)-284(is)-283(gi)25(v)15(en)-283(as)-283(an)-284(input,)]TJ/F23 9.9626 Tf 150.826 0 Td [(G)]TJ/F22 6.9738 Tf 7.833 -1.494 Td [(C)-70(AP)]TJ/F76 9.9626 Tf 21.659 1.494 Td [(generates)-283(lan-)]TJ -180.318 -11.955 Td [(guage)-315(hints)-315(by)-315(captioning)-315(the)-315(mask)10(ed)-315(image.)-505(Then,)-331(using)]TJ 0 -11.955 Td [(the)-204(language)-203(hint,)]TJ/F23 9.9626 Tf 72.307 0 Td [(G)]TJ/F22 6.9738 Tf 7.833 -1.495 Td [(I)-78(M)]TJ/F76 9.9626 Tf 14.944 1.495 Td [(generates)-204(images)-203(for)-204(the)-203(mask)10(ed)-204(re-)]TJ
0 g 0 G
0 g 0 G
ET
Q
Q
q
1 0 0 1 0 0 cm
/40RfOf7fn-XLxv3GuJac7Q Do
Q
endstream
endobj
21 0 obj
<< /CS /DeviceRGB /I true /S /Transparency /Type /Group >>
endobj
22 0 obj
<< /Font << /F22 48 0 R /F23 49 0 R /F67 50 0 R /F75 40 0 R /F76 41 0 R /F97 51 0 R >> /ProcSet [ /PDF /Text /ImageC ] /XObject << /40RfOf7fn-XLxv3GuJac7Q 52 0 R /Im2 53 0 R >> >>
endobj
23 0 obj
<< /Length 16679 >>
stream
q
q
0 g 0 G
0 g 0 G
BT
/F76 9.9626 Tf 50.112 710.037 Td [(gion.)-301(The)-223(tw)10(o)-223(parts)-223(of)-223(CEP)-223(module)-223(operate)-223(in)-223(a)-224(pair)-223(to)-223(con-)]TJ 0 -11.955 Td [(duct)-341(either)-340(image)-341(outpainting)-341(or)-340(wide-range)-341(image)-341(blend-)]TJ 0 -11.955 Td [(ing,)-345(the)-325(tw)10(o)-326(most)-326(challenging)-325(image)-326(completion)-326(tasks.)-537(In)]TJ 0 -11.955 Td [(the)-282(follo)25(wing)-282(sections,)-291(we)-282(describe)-282(the)-282(process)-282(of)-283(t)1(h)-1(e)-282(CEP)]TJ 0 -11.955 Td [(module)-302(in)-302(details.)-467(Then,)-315(we)-303(will)-302(e)15(xplain)-302(ho)25(w)-302(to)-302(apply)-303(the)]TJ 0 -11.956 Td [(CEP)-250(module)-250(to)-250(e)15(xtensi)25(v)15(e)-250(painting)-250(tasks.)]TJ/F75 10.9589 Tf 0 -21.235 Td [(3.1.)-250(Captioningbased)-250(Extensi)10(v)10(e)-250(P)10(ainting)-250(Module)]TJ/F76 9.9626 Tf 11.955 -18.369 Td [(T)80(ypically)65(,)-372(image)-347(completion)-347(for)-348(lar)18(ge)-347(unkno)25(wn)-347(area)-348(is)]TJ -11.955 -11.955 Td [(challenging)-312(due)-311(to)-312(the)-311(limited)-312(amount)-312(of)-311(neighbouring)-312(in-)]TJ 0 -11.956 Td [(formation.)-623(Therefore,)-380(the)-354(goal)-354(of)-354(generating)-355(hints)-354(during)]TJ 0 -11.955 Td [(e)15(xtensi)25(v)15(e)-334(painting)-334(is)-335(to)-334(pro)15(vide)-334(as)-334(detailed)-334(as)-335(possible)-334(in-)]TJ 0 -11.955 Td [(formation)-289(about)-288(the)-289(input)-288(image)-289([12].)-425(In)-289(order)-289(to)-288(meet)-289(the)]TJ 0 -11.955 Td [(goal,)-333(we)-317(adopt)-316(an)-317(image)-317(captioning)-316(model,)-334(whi)1(ch)-317(aims)-317(to)]TJ 0 -11.955 Td [(describe)-399(the)-399(contents)-398(of)-399(an)-399(image)-399(with)-399(natural)-399(language.)]TJ 0 -11.955 Td [(Then,)-277(using)-272(this)-271(caption)-272(as)-271(a)-272(hint,)-277(we)-271<026c6c>-272(in)-272(the)-271(missing)-272(re-)]TJ 0 -11.956 Td [(gion)-320(of)-321(the)-320(image)-321(using)-320(a)-321(te)15(xt-guided)-320(image)-321(manipulation)]TJ 0 -11.955 Td [(netw)10(ork.)-310(Ov)15(erall)-250(process)-250(is)-250(as)-250(follo)25(ws:)]TJ 11.955 -12.392 Td [(Let)]TJ/F23 9.9626 Tf 15.85 0 Td [(I)]TJ/F22 6.9738 Tf 4.379 -1.494 Td [(GT)]TJ/F76 9.9626 Tf 15.049 1.494 Td [(be)-258(the)-258(ground-truth)-258(image)-258(and)]TJ/F23 9.9626 Tf 123.523 0 Td [(M)]TJ/F76 9.9626 Tf 13.321 0 Td [(be)-258(the)-258(binary)]TJ -184.077 -11.955 Td [(image)-276(\(1)-276(for)-276(the)-276(missing)-276(re)15(gion)-276(and)-276(0)-277(for)-276(the)-276(background\).)]TJ 0 -11.955 Td [(Then,)-250(an)-250(incomplete)-250(image)-250(can)-250(be)-250(represented)-250(as:)]TJ
ET
/S/Span<>BDC
BT
/F23 9.9626 Tf 120.289 443.967 Td [(I)]TJ/F22 6.9738 Tf 4.379 -1.494 Td [(I)-78(C)]TJ/F20 9.9626 Tf 13.541 1.494 Td [(=)]TJ/F23 9.9626 Tf 10.516 0 Td [(I)]TJ/F22 6.9738 Tf 4.38 -1.494 Td [(GT)]TJ/F26 9.9626 Tf 14.692 1.494 Td [(\f)]TJ/F20 9.9626 Tf 9.963 0 Td [(\(1)]TJ/F26 9.9626 Tf 11.07 0 Td [<00>]TJ/F23 9.9626 Tf 9.962 0 Td [(M)]TJ/F20 9.9626 Tf 10.751 0 Td [(\))]TJ/F23 9.9626 Tf 3.875 0 Td [(;)]TJ
ET
EMC
BT
/F76 9.9626 Tf 274.746 443.967 Td [(\(1\))]TJ -224.634 -22.791 Td [(where)]TJ/F26 9.9626 Tf 27.792 0 Td [(\f)]TJ/F76 9.9626 Tf 11.201 0 Td [(denotes)-347(the)-346(Hadamard)-347(product.)-600(The)-346(image)-347(cap-)]TJ -38.993 -11.955 Td [(tioning)-268(netw)10(ork)-267(generates)-268(the)-268(te)15(xt)-268(caption)-267(used)-268(for)-268(the)-268(lan-)]TJ 0 -11.956 Td [(guage)-250(hint)]TJ
ET
/S/Span<>BDC
BT
/F23 9.9626 Tf 124.557 384 Td [(T)]TJ/F22 6.9738 Tf 5.822 -1.494 Td [(hint)]TJ/F20 9.9626 Tf 18.683 1.494 Td [(=)]TJ/F23 9.9626 Tf 10.516 0 Td [(G)]TJ/F22 6.9738 Tf 7.833 -1.494 Td [(C)-70(AP)]TJ/F20 9.9626 Tf 18.837 1.494 Td [(\()]TJ/F23 9.9626 Tf 3.874 0 Td [(I)]TJ/F22 6.9738 Tf 4.38 -1.494 Td [(I)-78(C)]TJ/F20 9.9626 Tf 10.773 1.494 Td [(\))]TJ/F23 9.9626 Tf 3.875 0 Td [(:)]TJ
ET
EMC
BT
/F76 9.9626 Tf 274.746 384 Td [(\(2\))]TJ -212.679 -19.679 Td [(After)-380(obtaining)]TJ/F23 9.9626 Tf 66.229 0 Td [(T)]TJ/F22 6.9738 Tf 5.822 -1.495 Td [(hint)]TJ/F76 9.9626 Tf 15.915 1.495 Td [(,)-412(te)15(xt-guided)-380(image)-380(manipulation)]TJ -99.921 -11.955 Td [(netw)10(ork)]TJ/F23 9.9626 Tf 34.678 0 Td [(G)]TJ/F22 6.9738 Tf 7.834 -1.495 Td [(I)-78(M)]TJ/F76 9.9626 Tf 15.047 1.495 Td [(generates)-214(a)-214(complete)-214(image.)-297(The)-214(mask)10(ed)-214(im-)]TJ -57.559 -11.956 Td [(age)]TJ/F23 9.9626 Tf 16.458 0 Td [(I)]TJ/F22 6.9738 Tf 4.379 -1.494 Td [(I)-78(C)]TJ/F76 9.9626 Tf 13.404 1.494 Td [(and)-264(the)-264(language)-264(hint)]TJ/F23 9.9626 Tf 88.547 0 Td [(T)]TJ/F22 6.9738 Tf 5.822 -1.494 Td [(hint)]TJ/F76 9.9626 Tf 18.545 1.494 Td [(were)-264(used)-264(as)-264(inputs)-264(as)]TJ -147.155 -11.955 Td [(follo)25(w:)]TJ
ET
/S/Span<>BDC
BT
/F23 9.9626 Tf 113.23 315.19 Td [(I)]TJ/F22 6.9738 Tf 4.38 -1.494 Td [(pr)-31(ed)]TJ/F20 9.9626 Tf 19.213 1.494 Td [(=)]TJ/F23 9.9626 Tf 10.516 0 Td [(G)]TJ/F22 6.9738 Tf 7.833 -1.494 Td [(I)-78(M)]TJ/F20 9.9626 Tf 14.577 1.494 Td [(\()]TJ/F23 9.9626 Tf 3.874 0 Td [(I)]TJ/F22 6.9738 Tf 4.379 -1.494 Td [(I)-78(C)]TJ/F23 9.9626 Tf 10.774 1.494 Td [(;)-167(T)]TJ/F22 6.9738 Tf 10.25 -1.494 Td [(hint)]TJ/F20 9.9626 Tf 15.915 1.494 Td [(\))]TJ/F23 9.9626 Tf 5.535 0 Td [(;)]TJ
ET
EMC
BT
/F76 9.9626 Tf 274.746 315.19 Td [(\(3\))]TJ -224.634 -19.243 Td [(where)]TJ/F23 9.9626 Tf 26.829 0 Td [(I)]TJ/F22 6.9738 Tf 4.38 -1.494 Td [(pr)-31(ed)]TJ/F76 9.9626 Tf 18.936 1.494 Td [(denotes)-250(the)-250<026e616c>-250(output)-250(result.)]TJ/F75 9.9626 Tf -50.145 -12.392 Td [(Image)-286(Captioning)-286(Netw)10(ork)]TJ/F76 9.9626 Tf 119.694 0 Td [(Image)-286(captioning)-286(is)-286(the)-286(most)]TJ -119.694 -11.955 Td [(popular)-203(task)-203(in)-203(the)-202(multi-modal)-203<02656c642e>-295(Ho)25(we)25(v)15(er)40(,)-212(e)15(xisting)-203(im-)]TJ 0 -11.955 Td [(age)-282(captioning)-281(models)-282(ha)20(v)15(e)-281(been)-282(trained)-282(only)-281(on)-282(complete)]TJ 0 -11.955 Td [(images.)-376(Therefore,)-278(the)15(y)-272(are)-272(incapable)-272(of)-273(generating)-272(proper)]TJ 0 -11.955 Td [(captions)-408(of)-409(the)-408(mask)10(ed)-408(images.)-785(T)80(o)-408(o)15(v)15(ercome)-408(this)-409(prob-)]TJ 0 -11.956 Td [(lem,)-260(we)-259(perform)-258(additional)-258(training)-258(with)-258(randomly)-259(mask)10(ed)]TJ 0 -11.955 Td [(dataset)]TJ/F23 9.9626 Tf 30.671 0 Td [(I)]TJ/F22 6.9738 Tf 4.379 -1.494 Td [(r)-31(and)]TJ/F20 9.9626 Tf 21.525 1.494 Td [(=)]TJ/F23 9.9626 Tf 11.469 0 Td [(I)]TJ/F22 6.9738 Tf 4.379 -1.494 Td [(GT)]TJ/F26 9.9626 Tf 15.074 1.494 Td [(\f)]TJ/F20 9.9626 Tf 10.344 0 Td [(\(1)]TJ/F26 9.9626 Tf 11.45 0 Td [<00>]TJ/F20 9.9626 Tf 14.059 2.519 Td [(~)]TJ/F23 9.9626 Tf -3.715 -2.519 Td [(M)]TJ/F20 9.9626 Tf 10.751 0 Td [(\))]TJ/F76 9.9626 Tf 3.875 0 Td [(,)-315(where)]TJ/F20 9.9626 Tf 36.683 2.519 Td [(~)]TJ/F23 9.9626 Tf -3.715 -2.519 Td [(M)]TJ/F76 9.9626 Tf 13.756 0 Td [(is)-302(the)-301(random)]TJ -180.985 -11.955 Td [(mask.)]TJ 11.955 -12.392 Td [(F)15(or)-320(i)1(mage)-320(captioning,)-337(we)-319(optimized)-320(a)-319(language)-320(model,)]TJ -11.955 -11.955 Td [(which)-225(w)10(as)-225(pre-trained)-226(o)15(v)15(er)-225(a)-225(lar)18(ge)-225(dataset,)-231(with)-225(self-critical)]TJ 0 -11.955 Td [(sequence)-319(training)-318(\(SCST\))-319(approach)-318([24,)-319(29,)-318(30].)-516(This)-319(ap-)]TJ 0 -11.955 Td [(proach)-253(is)-252(based)-253(on)-253(the)-253(REINFORCE)-253(al)1(gorithm)-253([31],)-254(where)]TJ 0 -11.955 Td [(the)-304(re)25(w)10(ard)-305(is)-304(set)-304(to)-304(the)-305(metric)-304(used)-304(at)-304(the)-305(test)-304(time.)-473(Gi)25(v)15(en)]TJ 0 -11.956 Td [(the)-270(sentence)]TJ/F23 9.9626 Tf 51.857 0 Td [(T)]TJ/F22 6.9738 Tf 7.206 3.616 Td [(s)]TJ -1.384 -6.437 Td [(hint)]TJ/F76 9.9626 Tf 18.607 2.821 Td [(sampled)-270(from)-270(the)-270(polic)15(y)-270(during)-270(training)]TJ -76.286 -11.955 Td [(and)-224(a)-225(captioning)-224(model)-225(with)-224(parameters)]TJ/F23 9.9626 Tf 160.616 0 Td [<12>]TJ/F76 9.9626 Tf 4.954 0 Td [(,)-230(we)-224(minimize)-225(the)]TJ -165.57 -11.955 Td [(ne)15(g)5(ati)25(v)15(e)-250(e)15(xpected)-250(re)25(w)10(ard:)]TJ
ET
/S/Span<>BDC
BT
/F23 9.9626 Tf 101.422 81 Td [(L)]TJ/F22 6.9738 Tf 6.78 -1.494 Td [(R)]TJ/F20 9.9626 Tf 6.535 1.494 Td [(\()]TJ/F23 9.9626 Tf 3.875 0 Td [<12>]TJ/F20 9.9626 Tf 4.953 0 Td [(\))-278(=)]TJ/F26 9.9626 Tf 17.158 0 Td [<00>]TJ/F75 9.9626 Tf 7.749 0 Td [(E)]TJ/F22 6.9738 Tf 6.645 -1.494 Td [(T)]TJ/F21 4.9813 Tf 5.778 2.402 Td [(s)]TJ -1.073 -4.814 Td [(hint)]TJ/F25 6.9738 Tf 14.474 2.412 Td [<18>]TJ/F22 6.9738 Tf 6.227 0 Td [(p)]TJ/F21 4.9813 Tf 4.107 -1.057 Td [<12>]TJ/F20 9.9626 Tf 6.202 2.551 Td [([)]TJ/F23 9.9626 Tf 2.767 0 Td [(r)]TJ/F20 9.9626 Tf 4.772 0 Td [(\()]TJ/F23 9.9626 Tf 3.874 0 Td [(T)]TJ/F22 6.9738 Tf 7.206 4.114 Td [(s)]TJ -1.384 -6.577 Td [(hint)]TJ/F20 9.9626 Tf 15.916 2.463 Td [(\)])]TJ/F23 9.9626 Tf 8.302 0 Td [(;)]TJ
ET
EMC
BT
/F76 9.9626 Tf 274.746 81 Td [(\(4\))]TJ
0 g 0 G
0 g 0 G
34.116 629.037 Td [(where)]TJ/F23 9.9626 Tf 28.327 0 Td [(r)]TJ/F76 9.9626 Tf 8.759 0 Td [(is)-400(the)-401(re)25(w)10(ard)-400(computed)-400(by)-401(an)-400(e)25(v)25(aluation)-400(metric)]TJ -37.086 -11.955 Td [(\(e.g.)-479(CIDEr\),)-320(by)-306(comparing)-307(the)-306(generated)-306(sequence)-306(to)-307(the)]TJ 0 -11.955 Td [(corresponding)-322(ground-truth)-321(sequences.)-525(Then,)-340(the)-322(gradient)]TJ 0 -11.955 Td [(of)-250(\(4\))-250(can)-250(be)-250(approximated)-250(by:)]TJ
ET
/S/Span<>BDC
BT
/F26 9.9626 Tf 313.715 640.032 Td [(r)]TJ/F22 6.9738 Tf 8.302 -1.494 Td [<12>]TJ/F23 9.9626 Tf 4.507 1.494 Td [(L)]TJ/F22 6.9738 Tf 6.78 -1.494 Td [(R)]TJ/F20 9.9626 Tf 6.536 1.494 Td [(\()]TJ/F23 9.9626 Tf 3.874 0 Td [<12>]TJ/F20 9.9626 Tf 4.954 0 Td [(\))]TJ/F26 9.9626 Tf 6.642 0 Td [<19>-278<00>]TJ/F13 9.9626 Tf 19.925 11.059 Td [<10>]TJ/F23 9.9626 Tf 5.95 -11.059 Td [(r)]TJ/F20 9.9626 Tf 4.771 0 Td [(\()]TJ/F23 9.9626 Tf 3.875 0 Td [(T)]TJ/F22 6.9738 Tf 7.205 4.114 Td [(s)]TJ -1.383 -6.577 Td [(hint)]TJ/F20 9.9626 Tf 15.915 2.463 Td [(\))]TJ/F26 9.9626 Tf 6.088 0 Td [<00>]TJ/F23 9.9626 Tf 9.963 0 Td [(r)]TJ/F20 9.9626 Tf 4.771 0 Td [(\()]TJ 5.817 2.518 Td [(^)]TJ/F23 9.9626 Tf -1.942 -2.518 Td [(T)]TJ/F22 6.9738 Tf 5.822 -1.494 Td [(hint)]TJ/F20 9.9626 Tf 15.915 1.494 Td [(\))]TJ/F13 9.9626 Tf 3.875 11.059 Td [<11>]TJ/F26 9.9626 Tf 7.61 -11.059 Td [(r)]TJ/F22 6.9738 Tf 8.302 -1.494 Td [<12>]TJ/F20 9.9626 Tf 6.168 1.494 Td [(log)]TJ/F23 9.9626 Tf 14.529 0 Td [(p)]TJ/F22 6.9738 Tf 5.012 -1.494 Td [<12>]TJ/F20 9.9626 Tf 4.508 1.494 Td [(\()]TJ/F23 9.9626 Tf 3.874 0 Td [(T)]TJ/F22 6.9738 Tf 7.206 4.114 Td [(s)]TJ -1.384 -6.577 Td [(hint)]TJ/F20 9.9626 Tf 15.915 2.463 Td [(\))]TJ/F23 9.9626 Tf 3.875 0 Td [(;)]TJ
ET
EMC
BT
/F76 9.9626 Tf 533.496 625.766 Td [(\(5\))]TJ -224.634 -11.956 Td [(where)]TJ/F23 9.9626 Tf 27.984 0 Td [(r)]TJ/F20 9.9626 Tf 4.771 0 Td [(\()]TJ 5.817 2.519 Td [(^)]TJ/F23 9.9626 Tf -1.942 -2.519 Td [(T)]TJ/F22 6.9738 Tf 5.822 -1.494 Td [(hint)]TJ/F20 9.9626 Tf 15.915 1.494 Td [(\))]TJ/F76 9.9626 Tf 7.52 0 Td [(is)-366(the)-366(baseline)-366(re)25(w)10(ard)-365(obtained)-366(by)-366(greed-)]TJ -65.887 -11.955 Td [(ily)-424(decoding)-424(the)-424(model.)-832(This)-424(gradient)-424(tends)-424(to)-424(increase)]TJ 0 -11.955 Td [(the)-473(probability)-473(of)-474(the)-473(caption)-473(sampling)-473(from)-473(the)-474(polic)15(y)]TJ 0 -11.955 Td [(during)-476(training)-476(than)-476(the)-476(re)25(w)10(ard)-476(from)-477(the)-476(current)-476(model)]TJ 0 -11.955 Td [([32].)-303(Through)-230(this)-230(process,)-234(our)-229(ne)25(wly)-230(optimized)-230(captioning)]TJ 0 -11.955 Td [(model)-289(is)-289(able)-289(to)-289(predict)-289(the)-289(appropriate)-289(caption,)-299(re)15(g)5(ardless)]TJ 0 -11.956 Td [(of)-250(the)-250(mask)-250(shapes.)]TJ 11.955 -16.738 Td [(One)-206(of)-205(the)-206(adv)25(antages)-206(of)-205(language)-206(hints)-205(o)15(v)15(er)-206(image)-206(hints)]TJ -11.955 -11.955 Td [(is)-422(that)-423(it)-422(w)10(orks)-423(well)-422(re)15(g)5(ardless)-422(of)-423(the)-422(image)-423(completion)]TJ 0 -11.956 Td [(tasks.)-352(Prior)-264(image)-264(hint-based)-264(methods)-264([8,)-264(9,)-265(10])-264(ha)20(v)15(e)-264(lar)18(ge)]TJ 0 -11.955 Td [(dependenc)15(y)-309(on)-309(image)-308(structure.)-487(More)-309(speci)1<0263616c6c79>65(,)-324(BR)-309([8])]TJ 0 -11.955 Td [(and)-255(MR)-256([9])-255(are)-256(applicable)-255(only)-256(to)-255(symmetric)-256(images,)-257(since)]TJ 0 -11.955 Td [(the)15(y)-310(either)-310(rearrange)-310(the)-309(input)-310(image)-310(by)-310(switching)-310(the)-310(left)]TJ 0 -11.955 Td [(and)-241(right)-242(side,)-243(or)-242(mirror)-241(the)-241(input)-242(image)-241(ne)15(xt)-242(to)-241(the)-242(mask)10(ed)]TJ 0 -11.955 Td [(re)15(gion.)-663(Moreo)15(v)15(er)40(,)-397(IAH)-367([10])-368(can)-367(generate)-368(only)-367<0278>15(ed)-368(size)]TJ 0 -11.956 Td [(hints.)-524(On)-322(the)-321(contrary)65(,)-340(since)-321(the)-321(CEP)-322(uses)-321(other)-322(modality)]TJ 0 -11.955 Td [(as)-294(the)-294(hint,)-305(it)-294(can)-293(be)-294(applied)-294(to)-294(an)15(y)-294(type)-294(of)-294(image)-294(comple-)]TJ 0 -11.955 Td [(tion)-250(without)-250(an)15(y)-250(restrictions.)]TJ/F75 9.9626 Tf 0 -16.739 Td [(T)92(ext-guided)-672(Image)-671(Manipulation)-672(Netw)10(ork)]TJ/F76 9.9626 Tf 198.99 0 Td [(Recently)65(,)]TJ -198.99 -11.955 Td [(thanks)-320(to)-320(the)-320(w)10(orks)-320(on)-320(massi)25(v)15(e)-320(te)15(xt-image)-320(paired)-320(datasets)]TJ 0 -11.955 Td [([15,)-371(33,)-370(34,)-371(35],)-401(lar)18(ge-scale)-370(language-vision)-371(models)-371(such)]TJ 0 -11.955 Td [(as)-513(GLIDE)-513([13],)-579(Imagen)-513([36],)-579(D)40(ALL-E)-513([36],)-579(and)-514(te)15(xt-)]TJ 0 -11.955 Td [(dri)25(v)15(e)-496(blended)-496(dif)25(fusion)-496([14])-496(ha)20(v)15(e)-496(achie)25(v)15(ed)-496(an)-496(unprece-)]TJ 0 -11.956 Td [(dented)-289(l)1(e)25(v)14(el)-288(of)-289(generaliza)1(tion.)-426(In)-289(other)-288(w)10(ords,)-298(these)-289(lar)18(ge-)]TJ 0 -11.955 Td [(scale)-479(models)-478(trained)-479(on)-478(lar)18(ge-scale)-479(datasets)-478(are)-479(compe-)]TJ 0 -11.955 Td [(tent)-327(in)-328(zero-shot)-327(image)-328(generation)-327(on)-328(a)-327(wide)-328(range)-327(of)-328(do-)]TJ 0 -11.955 Td [(mains,)-487(e)25(v)15(en)-439(compared)-440(to)-439(models)-439(trained)-440(on)-439(the)-440<73706563690263>]TJ 0 -11.955 Td [(dataset.)-406(Therefore,)-290(we)-282(simply)-282(utilize)-282(these)-282(models)-282(without)]TJ 0 -11.955 Td [(an)15(y)-226(change.)-301(W)80(e)-226(e)15(xpect)-225(these)-226<62656e65027473>-225(to)-226(enrich)-225(v)25(arious)-226(real-)]TJ 0 -11.956 Td [(life)-370(applications)-369(of)-370(e)15(xtensi)25(v)15(e)-370(painting.)-669(Ne)15(xt,)-399(we)-370(will)-370(ap-)]TJ 0 -11.955 Td [(ply)-249(our)-248(CEP)-249(module)-249(to)-248(the)-249(most)-248(dif)25<0263756c74>-249(image)-249(completion)]TJ 0 -11.955 Td [(tasks,)-250(image)-250(outpainting)-250(and)-250(wide-range)-250(image)-250(blending.)]TJ/F75 10.9589 Tf 0 -34.276 Td [(3.2.)-250(Image)-250(Outpainting)]TJ/F76 9.9626 Tf 11.955 -22.716 Td [(The)-273(goal)-274(of)-273(image)-274(outpainting)-273(is)-273(to)-274(generate)-273(the)-274(outside)]TJ -11.955 -11.955 Td [(of)-248(the)-248(images)-247(when)-248(the)-248(gi)25(v)15(en)-248(information)-247(is)-248(only)-248(the)-248(inside)]TJ 0 -11.955 Td [(of)-385(the)-386(images.)-716(Generally)65(,)-420(the)-385(outpainting)-386(task)-385(predicts)-386(a)]TJ 0 -11.955 Td [(single)-311(side)-311(or)-311(both)-311(sides)-311(of)-311(images)-311(in)-311(horizontal)-311(direction.)]TJ 0 -11.956 Td [(Since)-389(we)-390(optimized)-389(the)-390(image)-389(captioning)-390(netw)10(ork)-389(of)-390(the)]TJ 0 -11.955 Td [(CEP)-342(module)-341(for)-342(random)-341(mask)10(ed)-342(images,)-364(we)-342(can)-342(perform)]TJ 0 -11.955 Td [(the)-325(outpainting)-326(task)-325(by)-326(putting)-325(only)-326(the)-325(mask)-326(correspond-)]TJ 0 -11.955 Td [(ing)-362(to)-361(the)-362(missing)-362(re)15(gion)-361(and)-362(passing)-362(it)-361(through)-362(the)-362(CEP)]TJ 0 -11.955 Td [(module.)]TJ
0 g 0 G
0 g 0 G
ET
Q
Q
q
1 0 0 1 0 0 cm
/jdTkq3WTW0ZHHEfe_iX1PQ Do
Q
endstream
endobj
24 0 obj
<< /Font << /F13 54 0 R /F20 55 0 R /F21 56 0 R /F22 48 0 R /F23 49 0 R /F25 57 0 R /F26 58 0 R /F75 40 0 R /F76 41 0 R >> /ProcSet [ /PDF /Text ] /XObject << /jdTkq3WTW0ZHHEfe_iX1PQ 59 0 R >> >>
endobj
25 0 obj
<< /Length 10916 >>
stream
q
q
0 g 0 G
0 g 0 G
0 g 0 G
BT
/F76 8.9664 Tf 50.112 713.885 Td [(T)80(able)-229(1.)-230(Quantitati)25(v)15(e)-229(results)-229(for)-230(the)-229(image)-230(outpaitning)-229(task)-229(on)-230(Landscape)-229([6],)-234(Landma)1(rks)-230([37],)-233(and)-230(AmsterT)35(ime)-229([38])-229(dataset,)-234(using)-229(metrics)]TJ 0 -10.959 Td [(on)]TJ 12.001 1.955 Td []TJ -1 -1.955 Td [(Frenchet)-227(Inception)-227(Distance)-227(\(FID\))-226([39])-227(\(the)-227(lo)25(wer)40(,)-232(the)-227(better\))-227(and)-226(K)25(ernel)-227(Inception)-227(Distance)-227(\(KID\))-227([40])-227(\(the)-227(lo)25(wer)40(,)-231(the)-227(better\).)-302(The)-227(best)]TJ -11.001 -10.959 Td [(score)-250(for)-250(each)-250(column)-250(is)-250(depicted)-250(in)-250(bold)-250(letters.)]TJ
ET
1 0 0 1 136.749 633.466 cm
q
.95 0 0 .95 0 0 cm
q
1 0 0 1 2.491 50.71 cm
[]0 d 0 J 0.398 w 0 0 m 336.169 0 l S
Q
1 0 0 1 -136.749 -633.466 cm
BT
/F76 9.9626 Tf 168.455 675.608 Td [(Outpainting)-5445(Landsc)1(ape)-2513(Landmarks)-2141(AmsterT)35(ime)]TJ 8.304 -11.955 Td [(Method)-5885(FID)-1756(KID)-1757(FI)1(D)-1757(KID)-1756(FID)-1756(KID)]TJ
ET
q
1 0 0 1 139.24 659.867 cm
[]0 d 0 J 0.398 w 0 0 m 336.169 0 l S
Q
BT
/F76 9.9626 Tf 169.979 651.299 Td [(CEP)-250(\(ours\))]TJ/F75 9.9626 Tf 93.225 0 Td [(25.46)-1200(0.001)-1200(13.48)-1200(0.001)-1200(26.98)-1200(0.002)]TJ/F76 9.9626 Tf -98.759 -11.955 Td [(Boundless)-250([4])-4330(44.19)-1200(0.011)-1200(22.38)-1200(0.011)-1200(59.12)-1200(0.026)]TJ -19.227 -11.955 Td [(Image)-250(Outpainting)-250([41])-2400(75.79)-1200(0.038)-1200(62.67)-1200(0.055)-1200(84.91)-1200(0.086)]TJ 30.296 -11.955 Td [(SRN)-250([5])-5441(37.07)-1200(0.016)-1200(37.49)-1200(0.030)-1200(63.12)-1200(0.034)]TJ -8.573 -11.956 Td [(NS-OUT)-250([6])-4580(52.28)-1200(0.072)-1200(43.87)-1200(0.030)-1200(6)-1(6.84)-1200(0.038)]TJ 2.286 -11.955 Td [(P)15(alette)-250([42])-4810(64.66)-1200(0.038)-1200(15.31)-1200(0.004)-1200(39.22)-1200(0.009)]TJ
ET
q
1 0 0 1 139.24 587.737 cm
[]0 d 0 J 0.398 w 0 0 m 336.169 0 l S
Q
1 0 0 1 136.749 633.466 cm
Q
0 g 0 G
0 g 0 G
1 0 0 1 -136.749 -633.466 cm
BT
/F76 8.9664 Tf 50.112 571.765 Td [(T)80(able)-332(2.)-331(Quantitati)25(v)15(e)-332(results)-332(for)-331(wide-range)-332(image)-331(blending)-332(task)-332(on)-331(Landmarks)-332([37],)-352(Scenery)-332([6],)-352(and)-331(AmsterT)35(ime)-332([38])-332(datasets)-331(using)]TJ 0 -10.959 Td [(metrics)-311(on)-312(on)-311(FID)-312([39])-311(\(the)-312(lo)25(wer)40(,)-326(the)-312(better\),)-327(and)-311(KID)-312([40])-311(\(the)-312(lo)25(wer)40(,)-326(the)-312(better\).)-494(The)-312(best)-311(score)-312(for)-311(each)-311(column)-312(is)-311(depicted)-312(in)-311(bold)]TJ 0 -10.959 Td [(letters.)]TJ
ET
1 0 0 1 89.725 487.505 cm
q
.95 0 0 .95 0 0 cm
q
1 0 0 1 2.491 56.687 cm
[]0 d 0 J 0.398 w 0 0 m 435.167 0 l S
Q
1 0 0 1 -89.725 -487.505 cm
BT
/F76 9.9626 Tf 98.193 535.624 Td [(W)40(ide-Range)-250(Image)-250(Blending)-3251(Landmarks)-3540(Scenery)-3279(AmsterT)35(ime)-2905(Image)-250(4K)]TJ 45.729 -11.955 Td [(model)-7559(FID)-2006(KID)-2006(FID)-2006(KID)-2006(FID)-2006(KID)-2007(FI)1(D)-2007(KID)]TJ
ET
q
1 0 0 1 92.216 519.883 cm
[]0 d 0 J 0.398 w 0 0 m 435.167 0 l S
Q
BT
/F76 9.9626 Tf 134.098 511.316 Td [(CEP)-250(\(ours\))]TJ/F75 9.9626 Tf 106.859 0 Td [(20.39)-1450(0.004)]TJ/F76 9.9626 Tf 73.724 0 Td [(38.05)-1450(0.017)]TJ/F75 9.9626 Tf 73.723 0 Td [(29.43)-1450(0.003)-1450(29.41)-1450(0.003)]TJ/F76 9.9626 Tf -257.623 -11.956 Td [(BRIDGE)-250([7])-5921(36.72)-1450(0.022)]TJ/F75 9.9626 Tf 183.9 0 Td [(36.31)-1450(0.011)]TJ/F76 9.9626 Tf 73.723 0 Td [(71.95)-1450(0.047)-1450(40.65)-1450(0.009)]TJ -246.002 -11.955 Td [(CA)-250([3])-7087(52.49)-1451(0.016)-1450(91.87)-1450(0.074)-1450(43.46)-1450(0.021)-1450(61.46)-1450(0.028)]TJ -13.833 -11.955 Td [(PEN-Net)-250([43])-5699(79.57)-1450(0.035)-1200(159.70)-1200(0.115)-1450(74.67)-1450(0.047)-1450(86.71)-1450(0.049)]TJ 7.741 -11.955 Td [<4869026c6c>-250([44])-6476(70.90)-1450(0.041)-1200(139.39)-1200(0.123)-1450(74.82)-1450(0.058)-1450(76.57)-1450(0.042)]TJ 3.323 -11.955 Td [(SRN)-250([5])-6809(94.35)-1450(0.041)-1450(70.94)-1450(0.039)-1450(81.21)-1450(0.048)-1200(123)-1(.58)-1200(0.109)]TJ -8.573 -11.955 Td [(NS-OUT)-250([6])-5699(103.77)-1200(0.090)-1450(82.69)-1450(0.044)-1200(135.63)-1200(0.144)-1200(116.64)-1200(0.069)]TJ
ET
q
1 0 0 1 92.216 435.799 cm
[]0 d 0 J 0.398 w 0 0 m 435.167 0 l S
Q
1 0 0 1 89.725 487.505 cm
Q
0 g 0 G
1 0 0 1 -89.725 -487.505 cm
BT
/F75 10.9589 Tf 50.112 408.307 Td [(3.3.)-250(W)18(iderange)-250(Image)-250(Blending)]TJ/F76 9.9626 Tf 11.955 -18.615 Td [(T)80(o)-493(perform)-494(the)-493(wide-range)-494(image)-493(blending)-493(task,)-555(the)]TJ -11.955 -11.955 Td [(o)15(v)15(erall)-213(process)-214(consists)-213(of)-214(three)-213(stages.)-298(Stage)-213(1)-214(is)-213(the)-214(multi-)]TJ 0 -11.955 Td [(step)-418(prediction,)-460(which)-417(repeats)-418(outpainting)-418(using)-418(the)-418(pre-)]TJ 0 -11.956 Td [(dicted)-402(output)-402(from)-402(the)-401(pre)25(vious)-402(step)-402(as)-402(the)-402(input)-402(for)-402(the)]TJ 0 -11.955 Td [(ne)15(xt)-332(step.)-556(In)-332(stage)-332(2,)-353(outpainting)-332(is)-332(repeated)-332(in)-332(the)-332(oppo-)]TJ 0 -11.955 Td [(site)-348(direction)-347(to)-348(stage)-348(1)-348(until)-347(the)-348(tw)10(o)-348(e)15(xtrapolated)-348(images)]TJ 0 -11.955 Td [(become)-283(connected.)-410(Finally)65(,)-292(in)-283(stage)-284(3,)-291(we)-284(apply)-283(a)-283(mask)-284(to)]TJ 0 -11.955 Td [(the)-218(disconnected)-218(re)15(gion)-218(and)-217(pass)-218(the)-218(mask)10(ed)-218(image)-218(through)]TJ 0 -11.956 Td [(our)-250(CEP)-250(module)-250(once.)]TJ/F75 11.9552 Tf 0 -25.957 Td [(4.)-250(Experiments)]TJ/F75 10.9589 Tf 0 -19.611 Td [(4.1.)-250(Baseline)-250(methods)]TJ/F76 9.9626 Tf 11.955 -18.616 Td [(Our)-383(proposed)-383(CEP)-383(module)-383(is)-383(model-agnostic)-384(for)-383(both)]TJ -11.955 -11.955 Td [(image)-668(captioning)-669(and)-668(te)15(xt-guided)-668(image)-669(manipulation)]TJ 0 -11.955 Td [(tasks.)-308(F)15(or)-246(the)-245(image)-245(captioning)-246(task,)-246(we)-245(implemented)-246(OF)74(A)]TJ 0 -11.955 Td [([11])-329(and)-329(ClipCap)-330([12],)-349(and)-329(for)-329(the)-329(te)15(xt-guided)-329(image)-330(ma-)]TJ 0 -11.955 Td [(nipulation)-246(task,)-246(we)-246(utilized)-246(GLIDE)-245([13])-246(and)-246(the)-246(te)15(xt-dri)25(v)15(en)]TJ 0 -11.955 Td [(blended)-250(dif)25(fusion)-250(model)-250([14].)]TJ/F75 10.9589 Tf 0 -21.973 Td [(4.2.)-250(Datasets)]TJ/F76 9.9626 Tf 11.955 -18.615 Td [(W)80(e)-236(e)25(v)25(aluate)-235(our)-236(module)-236(on)-236(con)40(v)15(entional)-235(datasets)-236(for)-236(e)15(x-)]TJ -11.955 -11.955 Td [(tensi)25(v)15(e)-282(painting,)-290(Scenery6000)-281([6])-282(and)-282(Beach)-282(datasets)-282([41].)]TJ 0 -11.955 Td [(W)80(e)-432(further)-432(conduct)-432(e)15(xperiments)-432(on)-432(complicated)-432(datasets)]TJ 0 -11.956 Td [(such)-250(as)-250(AmsterT)35(ime)-250([38])-250(and)-250(Landmarks)-250([37])-250(datasets.)]TJ/F75 9.9626 Tf 0 -12.637 Td [(Beach)-348(dataset)]TJ/F76 9.9626 Tf 63.384 0 Td [([41])-348(This)-348(dataset)-348(w)10(as)-348(generated)-348(by)-349(select)1(-)]TJ
0 g 0 G
0 g 0 G
195.366 327.307 Td [(ing)-266(images)-266(from)-266('beach')-266(cate)15(gory)-266(of)-266(place365)-266([45].)-358(It)-266(con-)]TJ 0 -11.955 Td [(sists)-246(of)-247(9,465)-247(t)1(rain)-247(images)-246(and)-247(1,050)-247(test)-246(images,)-247(each)-247(with)]TJ/F20 9.9626 Tf 0 -11.955 Td [(256)]TJ/F26 9.9626 Tf 17.158 0 Td [<02>]TJ/F20 9.9626 Tf 9.962 0 Td [(256)]TJ/F76 9.9626 Tf 17.435 0 Td [(pix)15(el)-250(resolution.)]TJ/F75 9.9626 Tf -44.555 -12.236 Td [(Scenery6000)-253(dataset)]TJ/F76 9.9626 Tf 89.163 0 Td [([6])-253(This)-254(dataset)-253(contains)-254(scenery)-253(im-)]TJ -89.163 -11.955 Td [(ages)-383(of)-382(v)25(arying)-383(sizes.)-708(Among)-382(the)-383(6,040)-383(i)1(mages)-383(in)-383(total,)]TJ 0 -11.955 Td [(1,000)-250(are)-250(for)-250(testing,)-250(and)-250(5,040)-250(images)-250(are)-250(for)-250(training.)]TJ/F75 9.9626 Tf 0 -12.236 Td [(AmsterT)18(ime)-300(datase)1(t)]TJ/F76 9.9626 Tf 89.893 0 Td [([38])-300(This)-299(dataset)-300(includes)-299(1,231)-300(im-)]TJ -89.893 -11.955 Td [(ages)-235(of)-235(Amsterdam')55(s)-235(urbanscape,)-238(with)]TJ/F20 9.9626 Tf 155.889 0 Td [(800)]TJ/F26 9.9626 Tf 16.609 0 Td [<02>]TJ/F20 9.9626 Tf 9.414 0 Td [(600)]TJ/F76 9.9626 Tf 17.286 0 Td [(pix)15(el)-235(res-)]TJ -199.198 -11.955 Td [(olution.)]TJ/F75 9.9626 Tf 0 -12.236 Td [(Landmarks)-194(Dataset)]TJ/F76 9.9626 Tf 85.769 0 Td [([37])-193(This)-194(dataset)-194(contains)-193(26,397)-194(train)]TJ -85.769 -11.955 Td [(images)-496(and)-496(3,103)-496(test)-495(images)-496(of)-496(landmarks)-496(around)-496(the)]TJ 0 -11.956 Td [(w)10(orld,)-250(ranging)-250(from)-250(natural)-250(landscapes)-250(to)-250(architectures.)]TJ/F75 9.9626 Tf 0 -12.235 Td [(4K)-267(datas)1(et)]TJ/F76 9.9626 Tf 48.478 0 Td [(This)-267(is)-266(a)-267(Kaggle)-266(pro)15(vided)-267(dataset)-266(composed)-267(of)]TJ -48.478 -11.955 Td [(v)25(arious)-242(scenery)-242(images)-242(and)-242(object)-242(images)-243(in)-242(4K)-242(resolution.)]TJ 0 -11.956 Td [(All)-250(of)-250(the)-250(2,056)-250(images)-250(were)-250(used)-250(for)-250(testing.)]TJ/F75 10.9589 Tf 0 -20.767 Td [(4.3.)-250(System)-250(Setup)]TJ/F76 9.9626 Tf 11.955 -18.213 Td [(T)80(o)-508(optimize)-508(OF)74(A)-508([11])-508(via)-508(SCST)-508(optimization)-508(using)]TJ -11.955 -11.955 Td [(CIDEr)-425(e)25(v)25(aluation)-424(metric,)-468(we)-425(utilize)-425(the)-424(most)-425(commonly)]TJ 0 -11.955 Td [(used)-279(MS)-280(COCO)-279(Caption)-279(dataset)-280([46].)-398(Opimization)-279(is)-280(con-)]TJ 0 -11.955 Td [(ducted)-266(for)-266(5,000)-266(steps)-266(with)-265(a)-266(batch)-266(size)-266(of)-266(2)-266(and)-266(a)-266(learning)]TJ 0 -11.956 Td [(rate)-284(of)-284(5e-6.)-412(As)-284(for)-284(the)-284(other)-285(image)-284(captioning)-284(model)-284([12])]TJ 0 -11.955 Td [(and)-385(te)15(xt-guided)-386(image)-385(manipulation)-385(models)-385([13,)-386(14],)-419(the)]TJ 0 -11.955 Td [(original)-250(setup)-250(is)-250(used.)]TJ 11.955 -12.236 Td [(F)15(or)-438(image)-438(outpainting,)-486(input)-438(images)-438(are)-438(cropped)-439(and)]TJ -11.955 -11.955 Td [(resized)-446(into)-447(the)-446(resolution)-447(of)]TJ/F20 9.9626 Tf 126.277 0 Td [(256)]TJ/F26 9.9626 Tf 18.607 0 Td [<02>]TJ/F20 9.9626 Tf 11.412 0 Td [(256)]TJ/F76 9.9626 Tf 14.944 0 Td [(,)-496(and)-446(for)-446(wide-)]TJ -171.24 -11.955 Td [(range)-367(image)-368(blending,)-396(input)-368(im)1(ages)-368(are)-367<02727374>-367(cropped)-368(and)]TJ 0 -11.955 Td [(resized)-402(into)-401(the)-402(resolution)-401(of)]TJ/F20 9.9626 Tf 124.043 0 Td [(768)]TJ/F26 9.9626 Tf 18.277 0 Td [<02>]TJ/F20 9.9626 Tf 11.081 0 Td [(256)]TJ/F76 9.9626 Tf 14.944 0 Td [(.)-765(Then)-401(for)-402(self-)]TJ
0 g 0 G
0 g 0 G
ET
Q
Q
q
1 0 0 1 0 0 cm
/K20d1HqapHSQe45t7R0CzA Do
Q
endstream
endobj
26 0 obj
<< /Font << /F20 55 0 R /F26 58 0 R /F75 40 0 R /F76 41 0 R >> /ProcSet [ /PDF /Text ] /XObject << /K20d1HqapHSQe45t7R0CzA 60 0 R >> >>
endobj
27 0 obj
<< /Length 4500 >>
stream
q
q
0 g 0 G
0 g 0 G
0 g 0 G
1 0 0 1 50.112 304.944 cm
q
.51563 0 0 .51563 0 0 cm
q
960 0 0 804.96 0 0 cm
/Im3 Do
Q
Q
1 0 0 1 -50.112 -304.944 cm
BT
/F76 8.9664 Tf 155.6 292.988 Td [(Figure)-250(3.)-560(Qualitati)25(v)15(e)-250(results)-250(of)-250(image)-250(outpainting)-250(task)-250(on)-250(AmsterT)35(ime)-250(dataset.)]TJ
0 g 0 G
/F76 9.9626 Tf -105.488 -28.994 Td [(reconstruction,)-230(the)-225(middle)]TJ/F20 9.9626 Tf 106.108 0 Td [(256)]TJ/F26 9.9626 Tf 16.232 0 Td [<02>]TJ/F20 9.9626 Tf 9.037 0 Td [(256)]TJ/F76 9.9626 Tf 17.185 0 Td [(re)15(gion)-225(is)-225(cropped,)-230(and)]TJ -148.562 -11.955 Td [(the)-315(remaining)-315(left)]TJ/F20 9.9626 Tf 75.266 0 Td [(256)]TJ/F26 9.9626 Tf 17.637 0 Td [<02>]TJ/F20 9.9626 Tf 10.442 0 Td [(256)]TJ/F76 9.9626 Tf 18.082 0 Td [(and)-315(right)]TJ/F20 9.9626 Tf 39.481 0 Td [(256)]TJ/F26 9.9626 Tf 17.638 0 Td [<02>]TJ/F20 9.9626 Tf 10.442 0 Td [(256)]TJ/F76 9.9626 Tf 18.082 0 Td [(re)15(gions)]TJ -207.07 -11.955 Td [(serv)15(e)-318(as)-319(the)-318(tw)10(o)-319(input)-318(images.)-516(The)-318(mask)-319(shape)-318(for)-319(image)]TJ 0 -11.956 Td [(outpainting)-291(is)]TJ/F20 9.9626 Tf 57.84 0 Td [(128)]TJ/F26 9.9626 Tf 17.464 0 Td [<02>]TJ/F20 9.9626 Tf 10.268 0 Td [(256)]TJ/F76 9.9626 Tf 14.944 0 Td [(.)-434(F)15(or)-292(wide-range)-291(image)-291(blending,)]TJ -100.516 -11.955 Td [(the)-280(mask)-280(shape)-280(for)-280(e)15(xtrapolation)-280(step)-281(is)]TJ/F20 9.9626 Tf 162.158 0 Td [(128)]TJ/F26 9.9626 Tf 17.38 0 Td [<02>]TJ/F20 9.9626 Tf 10.184 0 Td [(256)]TJ/F76 9.9626 Tf 17.735 0 Td [(and)-280(for)]TJ -207.457 -11.955 Td [(inpainting)-288(step,)]TJ/F20 9.9626 Tf 64.775 0 Td [(64)]TJ/F26 9.9626 Tf 12.454 0 Td [<02>]TJ/F20 9.9626 Tf 10.241 0 Td [(256)]TJ/F76 9.9626 Tf 14.944 0 Td [(.)-423(All)-288(models)-287(are)-288(implemented)-288(on)]TJ -102.414 -11.955 Td [(4)-250(NVID)40(A)-250(GeF)15(orce)-250(R)60(TX)-250(3090)-250(GPUs.)]TJ/F75 10.9589 Tf 0 -21.181 Td [(4.4.)-250(Quantitati)10(v)10(e)-250(Results)]TJ/F76 9.9626 Tf 11.955 -18.351 Td [(W)80(e)-459(used)-458(Frenchet)-459(Inception)-459(Distance)-459(\()1(FID\))-459([39])-459(and)]TJ -11.955 -11.955 Td [(K)25(ernel)-479(Inception)-478(Distance)-479(\(KID\))-479([40])-478(as)-479(our)-479(e)25(v)25(aluation)]TJ 0 -11.955 Td [(metrics.)-1010(Not)1(e)-484(that)-483(we)-483(ha)20(v)15(e)-483(utilized)-483(weights)-484(for)-483(Land-)]TJ 0 -11.956 Td [(marks)-349(dataset)-350(to)-349(test)-349(on)-350(AmsterT)35(ime)-349(dataset,)-374(for)-349(their)-350(vi-)]TJ 0 -11.955 Td [(sual)-236(similarities)-236(and)-237(the)-236(lack)-236(of)-236(training)-236(images)-237(for)-236(Amster)20(-)]TJ 0 -11.955 Td [(T)35(ime)-327(\(AmsterT)35(i)1(me)-327(dataset)-327(contains)-326(1,231)-327(images)-326(in)-327(total,)]TJ 0 -11.955 Td [(whereas)-383(Landmarks)-384(dataset)-383(possesses)-384(26,397)-383(images)-384(just)]TJ
0 g 0 G
0 g 0 G
258.75 182.994 Td [(for)-250(training\).)]TJ/F75 9.9626 Tf 0 -31.721 Td [(Image)-269(Outpainting)]TJ/F76 9.9626 Tf 83.958 0 Td [(The)-269(results)-268(for)-269(image)-269(outpainting)-268(task)]TJ -83.958 -11.955 Td [(on)-417(Landscape,)-460(Landmarks,)-459(and)-417(AmsterT)35(ime)-417(datasets)-418(are)]TJ 0 -11.956 Td [(summarized)-426(in)-427(T)80(able)-853(1.)-839(Our)-426(CEP)-426(module)-427(outperforms)]TJ 0 -11.955 Td [(all)-352(baselines,)-377(i)1(ncluding)-352(the)-352(state-of-the-art)-351(image)-352(comple-)]TJ 0 -11.955 Td [(tion)-270(model)-269(P)15(alette)-270([42].)-368(Moreo)15(v)15(er)39(,)-274(while)-270(the)-269(P)15(alette)-270(model)]TJ 0 -11.955 Td [(scores)-260(particul)1(arly)-260(high)-260(in)-259(Landmarks)-260(dataset,)-262(in)-259(which)-260(the)]TJ 0 -11.955 Td [(number)-315(of)-316(training)-315(images)-316(is)-315(lar)18(ge,)-332(our)-315(module)-316(constantly)]TJ 0 -11.955 Td [(achie)25(v)15(es)-250(high)-250(scores)-250(in)-250(all)-250(datasets.)]TJ/F75 9.9626 Tf 0 -31.722 Td [(W)18(ide-Range)-305(Image)-305(Blending)]TJ/F76 9.9626 Tf 126.827 0 Td [(The)-305(quantitati)25(v)15(e)-305(results)-304(are)]TJ -126.827 -11.955 Td [(sho)25(wn)-232(in)-233(T)80(able)-232(2.)-304(In)-232(all)-233(datasets)-232(e)15(xcept)-232(for)-232(Scenery)-233(dataset,)]TJ 0 -11.955 Td [(our)-289(method)-290(beats)-289(e)25(v)15(ery)-290(baseline,)-299(e)25(v)15(en)-290(the)-289(BRIDGE)-290(model)]TJ 0 -11.955 Td [([7],)-250(which)-250(w)10(as)-250(particularly)-250(designed)-250(for)-250(this)-250(task.)]TJ
0 g 0 G
0 g 0 G
ET
Q
Q
q
1 0 0 1 0 0 cm
/n995eAGEWWZlJ0P_Twcisw Do
Q
endstream
endobj
28 0 obj
<< /Font << /F20 55 0 R /F26 58 0 R /F75 40 0 R /F76 41 0 R >> /ProcSet [ /PDF /Text /ImageC ] /XObject << /Im3 61 0 R /n995eAGEWWZlJ0P_Twcisw 62 0 R >> >>
endobj
29 0 obj
<< /Length 5066 >>
stream
q
q
0 g 0 G
0 g 0 G
0 g 0 G
1 0 0 1 50.112 430.077 cm
q
.51537 0 0 .51537 0 0 cm
q
960.48 0 0 562.56 0 0 cm
/Im4 Do
Q
Q
1 0 0 1 -50.112 -430.077 cm
BT
/F76 8.9664 Tf 143.751 418.121 Td [(Figure)-250(4.)-560(Qualitati)25(v)15(e)-250(results)-250(of)-250(wide-range)-250(image)-250(blending)-250(task)-250(on)-250(Image)-250(4K)-250(dataset.)]TJ
0 g 0 G
/F75 10.9589 Tf -93.639 -28.994 Td [(4.5.)-250(Qualitati)10(v)10(e)-250(Results)]TJ/F76 9.9626 Tf 11.955 -18.146 Td [(Fig.)-299(3)-218(sho)25(ws)-218(qualitati)25(v)15(e)-218(results)-218(of)-218(con)40(v)15(entional)-218(outpaint-)]TJ -11.955 -11.955 Td [(ing)-201(algorithms)-202(and)-201(the)-201(CEP)-202(module)-201(on)-201(AmsterT)35(ime)-202(dataset.)]TJ 0 -11.955 Td [(Pre)25(vious)-201(methods)-201(tend)-201(to)-201(produce)-202(structural)1(ly)-202(unnatural)-201(im-)]TJ 0 -11.955 Td [(ages,)-473(b)20(ut)-429(our)-428(method)-428(produces)-429(structurally)-428(coherent)-429(and)]TJ 0 -11.955 Td [(content-preserving)-360(images.)-640(This)-360(characteristic)-360(also)-360(main-)]TJ 0 -11.956 Td [(tains)-248(for)-249(wide-range)-248(image)-249(blending)-248(task.)-309(As)-249(sho)25(wn)-248(in)-249(Fig.)]TJ 0 -11.955 Td [(4,)-403(our)-372(method)-372(produces)-372(clean)-372(and)-372(semantically)-372(meaning-)]TJ 0 -11.955 Td [(ful)-376(images)-376(while)-377(BRIDGE,)-376(the)-376(state-of-the-art)-376(method)-377(in)]TJ 0 -11.955 Td [(wide-range)-305(image)-304(blending,)-319(produces)-304(repetiti)25(v)15(e)-305(structures.)]TJ 0 -11.955 Td [(Additional)-250(results)-250(are)-250(to)-250(be)-250(found)-250(in)-250(the)-250(appendix.)]TJ/F75 10.9589 Tf 0 -20.564 Td [(4.6.)-250(Ablation)-250(Studies)]TJ/F75 9.9626 Tf 0 -18.146 Td [(Effect)-364(of)-364(optimizing)-364(an)-364(image)-365(capt)1(ioning)-365(model)]TJ/F76 9.9626 Tf 211.354 0 Td [(Image)]TJ -211.354 -11.955 Td [(captioning)-351(models)-352(created)-351(inaccurate)-351<7072650278>15(es)-351(when)-352(gi)25(v)15(en)]TJ 0 -11.955 Td [(mask)10(ed)-359(images,)-387(such)-359(as)-360<937477>10(o)-359(photos)-360<6f6694>-359(and)-359(its)-360(v)25(ariants.)]TJ 0 -11.956 Td [(When)-428(te)15(xt-guided)-427(image)-428(manipulation)-427(models)-428(were)-428(pro-)]TJ 0 -11.955 Td [(vided)-249(with)-249(such)-250(inaccurate)-249(captions,)-249(the)15(y)-249(generated)-249(tw)10(o)-250(di-)]TJ 0 -11.955 Td [(vided)-312(images)-313(during)-312(wide-range)-313(image)-312(blending)-312(as)-313(sho)25(wn)]TJ 0 -11.955 Td [(in)-198(Fig.)-293(5.)-293(Thus,)-209(we)-198(measured)-198(the)-198(ef)25(fect)-199(of)-198(optimizing)-198(an)-199(im-)]TJ 0 -11.955 Td [(age)-278(captioning)-278(model,)-286(OF)74(A,)-278(using)-278(SCST)-278(method)-278(with)-279(ran-)]TJ 0 -11.955 Td [(dom)-232(masks)1(.)-304(Then,)-236(we)-231(counted)-232(the)-231(number)-232(of)-231(f)10(alse)-232<7072650278>15(es.)]TJ 0 -11.956 Td [(As)-236(summarized)-236(in)-236(T)80(able)-237(3,)-239(S)1(CST)-237(optimization)-236(successfully)]TJ 0 -11.955 Td [(reduces)-250(the)-250(number)-250(of)-250(error)-250(cases.)]TJ/F75 9.9626 Tf 0 -12.168 Td [(Effect)-379(of)-379(captions)-379(f)25(or)-379(extensi)10(v)10(e)-379(painting)]TJ/F76 9.9626 Tf 178.267 0 Td [(In)-379(Fig.)-697(6,)-411(we)]TJ -178.267 -11.955 Td [(closely)-258(look)-259(at)-258(the)-259(ef)25(fect)-258(of)-258(captions)-259(for)-258(e)15(xtensi)25(v)15(e)-259(painting.)]TJ
0 g 0 G
0 g 0 G
0 g 0 G
ET
1 0 0 1 320.675 237.641 cm
q
.44312 0 0 .44312 0 0 cm
q
479.8389 0 0 364.349 0 0 cm
/Im5 Do
Q
Q
1 0 0 1 -320.675 -237.641 cm
BT
/F76 8.9664 Tf 308.862 226.683 Td [(Figure)-477(5.)-477(Ef)25(fect)-477(of)-477(optimizing)-477(an)-477(image)-477(captioning)-476(model)-477(on)]TJ 0 -10.959 Td [(wide-range)-336(image)-335(blending;)-379(W)40(ithoutthe)-335(SCST)-336(optimization)-335(pro-)]TJ 0 -10.959 Td [(cess,)-246(the)-245(te)15(xt-guided)-245(image)-245(manipulation)-245(model)-245(generated)-244(tw)10(o)-245(di-)]TJ 0 -10.959 Td [(vided)-231(images,)-234(whereas)-231(with)-230(the)-231(optimization)-230(process,)-235(we)-230(obtained)]TJ 0 -10.959 Td [(smoothly)-250(blended)-250(images)]TJ
0 g 0 G
/F76 9.9626 Tf 0 -30.116 Td [(In)-324(the)-323(image)-324(outpainting)-324(t)1(ask,)-343(a)-323(canal)-324(w)10(as)-323(included)-324(in)-324(the)]TJ 0 -11.955 Td [(generated)-347(image,)-372(because)-347(of)-347(the)-347(w)10(ord)-347<9363616e616c94>-347(in)-348(the)-347(cap-)]TJ 0 -11.955 Td [(tion.)-601(Note)-347(that)-347(this)-347(w)10(as)-347(not)-347(included)-347(in)-347(the)-347(initial)-347(image.)]TJ 0 -11.955 Td [(Also)-360(in)-361(the)-360(image)-361(blending)-360(task,)-388(a)-361(yello)25(w)-360(beak)-361(of)-360(an)-361(ea-)]TJ 0 -11.956 Td [(gle)-311(w)10(as)-311(generated)-311(through)-311(the)-311(caption,)-327(\223eagle)-311(with)-311(yello)25(w)]TJ 0 -11.955 Td [(beak\224.)-632(This)-358(indicat)1(es)-358(that)-357(image)-358(captions)-357(for)-357(missing)-358(re-)]TJ 0 -11.955 Td [(gions)-250(generate)-250(di)25(v)15(erse)-250(and)-250(natural)-250(images.)]TJ
0 g 0 G
0 g 0 G
ET
Q
Q
q
1 0 0 1 0 0 cm
/Zmkmzwcm5qOjquJ4fbYlFQ Do
Q
endstream
endobj
30 0 obj
<< /Font << /F75 40 0 R /F76 41 0 R >> /ProcSet [ /PDF /Text /ImageC ] /XObject << /Im4 63 0 R /Im5 64 0 R /Zmkmzwcm5qOjquJ4fbYlFQ 65 0 R >> >>
endobj
31 0 obj
<< /Length 9939 >>
stream
q
q
0 g 0 G
0 g 0 G
0 g 0 G
1 0 0 1 50.112 634.826 cm
q
.49236 0 0 .49236 0 0 cm
q
479.8389 0 0 172.9933 0 0 cm
/Im6 Do
Q
Q
1 0 0 1 -50.112 -634.826 cm
BT
/F76 8.9664 Tf 50.112 623.867 Td [(Figure)-236(6.)-236(Ef)25(fect)-236(of)-235(captions)-236(for)-236(e)15(xtensi)25(v)15(e)-236(painting;)-241(Image)-235(captions)]TJ 0 -10.959 Td [(for)-215(missing)-215(re)15(gion)-216(pro)15(vides)-215(di)25(v)15(erse)-215(and)-216(natural)-215(images.)-298(Red)-215(box)15(es)]TJ 0 -10.959 Td [(indicate)-250(generated)-250(re)15(gions.)]TJ
0 g 0 G
0 g 0 G
0 -24.765 Td [(T)80(able)-424(3.)-424(Ef)25(fect)-424(of)-424(SCST)-424(optimization)-424(on)-424(the)-424(image)-424(captioning)]TJ 0 -10.959 Td [(model;)-601(Image)-484(captioning)-485(model)-484(OF)74(A)-484([11])-484(w)10(as)-484(optimized)-484(via)]TJ 0 -10.958 Td [(SCST)-218(method)-219(using)-218(CIDEr)-219(with)-218(randomly)-219(mask)10(ed)-218(COCO)-218(caption)]TJ 0 -10.959 Td [(dataset)-314([46].)-504(GLDIE)-314([13])-314(w)10(as)-315(used)-314(for)-315(image)-314(te)15(xt-guided)-314(image)]TJ 0 -10.959 Td [(manipulation,)-241(and)-239(e)25(v)25(aluation)-239(w)10(as)-238(conducted)-239(on)-239(image)-239(4K)-238(dataset.)]TJ
ET
1 0 0 1 63.338 503.241 cm
q
.95 0 0 .95 0 0 cm
q
1 0 0 1 2.491 20.822 cm
[]0 d 0 J 0.398 w 0 0 m 218.35 0 l S
Q
1 0 0 1 -63.338 -503.241 cm
BT
/F76 9.9626 Tf 71.806 515.495 Td [(OF)74(A)-250(training)-1200(#)-250(of)-250(F)15(alse)-250<7072650278>15(es)-1519(FID)-1756(KID)]TJ
ET
q
1 0 0 1 65.829 511.709 cm
[]0 d 0 J 0.398 w 0 0 m 218.35 0 l S
Q
BT
/F76 9.9626 Tf 73.928 503.141 Td [(w/o)-250(training)-4078(1055)-3865(33.81)-1200(0.003)]TJ 2.491 -11.955 Td [(w/)-250(training)]TJ/F75 9.9626 Tf 94.036 0 Td [(3)-4615(32.55)-1200(0.002)]TJ
ET
q
1 0 0 1 65.829 487.4 cm
[]0 d 0 J 0.398 w 0 0 m 218.35 0 l S
Q
1 0 0 1 63.338 503.241 cm
Q
0 g 0 G
0 g 0 G
1 0 0 1 -63.338 -503.241 cm
BT
/F76 8.9664 Tf 50.112 462.345 Td [(T)80(able)-287(4.)-286(Ef)25(fect)-287(of)-287(mask)-287(sizes;)-305(Image)-286(inpainting)-287(w)10(as)-287(conducted)-286(at)]TJ 0 -10.959 Td [(the)-237(last)-237(stage)-236(of)-237(wide-range)-237(image)-237(blending)-237(on)-237(Landmarks)-236(dataset)]TJ 0 -10.959 Td [([37],)-250(using)-250(OF)74(A)-250([11])-250(and)-250(GLIDE)-250([13])-250(with)-250(v)25(arying)-250(mask)-250(widths.)]TJ
ET
1 0 0 1 110.192 398.962 cm
q
.95 0 0 .95 0 0 cm
q
1 0 0 1 2.49 32.777 cm
[]0 d 0 J 0.398 w 0 0 m 119.711 0 l S
Q
1 0 0 1 -110.192 -398.962 cm
BT
/F76 9.9626 Tf 118.66 423.171 Td [(mask)-250(size)-1519(FID)-1756(KID)]TJ
ET
q
1 0 0 1 112.682 419.385 cm
[]0 d 0 J 0.398 w 0 0 m 119.711 0 l S
Q
BT
/F76 9.9626 Tf 133.185 410.817 Td [(32)-2658(21.15)-1200(0.004)]TJ 0 -11.955 Td [(64)]TJ/F75 9.9626 Tf 36.443 0 Td [(20.85)-1200(0.003)]TJ/F76 9.9626 Tf -36.443 -11.955 Td [(96)-2658(20.90)-1200(0.003)]TJ -2.491 -11.955 Td [(128)-2408(22.12)-1200(0.004)]TJ
ET
q
1 0 0 1 112.682 371.166 cm
[]0 d 0 J 0.398 w 0 0 m 119.711 0 l S
Q
1 0 0 1 110.192 398.962 cm
Q
0 g 0 G
0 g 0 G
1 0 0 1 -110.192 -398.962 cm
BT
/F76 8.9664 Tf 50.112 346.708 Td [(T)80(able)-328(5.)-328(Comparison)-328(of)-328(hint-based)-328(image)-328(outpainting)-328(models)-328(on)]TJ 0 -10.959 Td [(beach)-250(dataset.)]TJ
ET
1 0 0 1 95.011 296.122 cm
q
.95 0 0 .95 0 0 cm
q
1 0 0 1 2.49 32.777 cm
[]0 d 0 J 0.398 w 0 0 m 151.67 0 l S
Q
1 0 0 1 -95.011 -296.122 cm
BT
/F76 9.9626 Tf 110.258 320.331 Td [(Method)-1881(Hint)-250(F)15(ormat)-1519(FID)]TJ
ET
q
1 0 0 1 97.501 316.545 cm
[]0 d 0 J 0.398 w 0 0 m 151.67 0 l S
Q
BT
/F76 9.9626 Tf 103.479 307.977 Td [(CEP)-250(\(ours\))-1679(Language)]TJ/F75 9.9626 Tf 117.299 0 Td [(24.95)]TJ/F76 9.9626 Tf -108.721 -11.955 Td [(BR)-250([8])-3262(Image)-2402(37.19)]TJ -1.106 -11.955 Td [(MR)-250([9])-3151(Image)-2402(36.65)]TJ -3.592 -11.955 Td [(IAH)-250([10])-2791(Image)-2401(31.81)]TJ
ET
q
1 0 0 1 97.501 268.326 cm
[]0 d 0 J 0.398 w 0 0 m 151.67 0 l S
Q
1 0 0 1 95.011 296.122 cm
Q
0 g 0 G
1 0 0 1 -95.011 -296.122 cm
BT
/F75 9.9626 Tf 50.112 232.05 Td [(Effect)-227(of)-228(Mask)-227(Sizes)]TJ/F76 9.9626 Tf 87.636 0 Td [(In)-227(the)-228<026e616c>-227(step)-227(of)-228(wide-image)-227(blend-)]TJ -87.636 -11.955 Td [(ing,)-241(we)-238(mask)10(ed)-238(the)-238(v)15(ery)-238(center)-238(of)-238(the)-238(panoramic)-239(image)-238(and)]TJ 0 -11.955 Td [(performed)-304(image)-305(inpainting)-304(for)-304(a)-305(s)1(mooth)-305(transition.)-473(From)]TJ 0 -11.955 Td [(T)80(able)-263(4,)-265(we)-263(can)-262(conclude)-263(that)-263(performing)-262(image)-263(inpainting)]TJ 0 -11.955 Td [(using)-250(a)-250(mask)-250(with)-250(a)-250(width)-250(of)-250(64,)-250(yields)-250(the)-250(best)-250(results.)]TJ/F75 9.9626 Tf 0 -15.75 Td [(Comparison)-447(on)-448(hint-based)-447(methods)]TJ/F76 9.9626 Tf 161.744 0 Td [(W)80(e)-447(compared)-448(our)]TJ -161.744 -11.955 Td [(method)-490(with)-489(con)40(v)15(entional)-490(hint-based)-489(methods)-490(on)-490(beach)]TJ 0 -11.955 Td [(dataset.)-553(As)-331(sho)25(wn)-331(in)-331(T)80(able)-331(5,)-351(our)-331(algorithm)-331(w)10(as)-331(superior)]TJ 0 -11.955 Td [(to)-250(the)-250(e)15(xisting)-250(ones.)]TJ/F75 9.9626 Tf 0 -15.75 Td [(Comparison)-290(on)-290(captioning)-290(models)]TJ/F76 9.9626 Tf 149.394 0 Td [(Finally)65(,)-300(we)-290(compared)]TJ -149.394 -11.955 Td [(the)-221(results)-221(of)-221(our)-221(method)-221(using)-222(dif)25(ferent)-221(captioning)-221(models.)]TJ 0 -11.955 Td [(In)-386(T)80(able)-386(6,)-420(the)-386(CEP)-386(module)-387(with)-386(OF)74(A)-386(outperformed)-386(the)]TJ 0 -11.955 Td [(CEP)-250(module)-250(with)-250(ClipCap.)]TJ
0 g 0 G
0 g 0 G
0 g 0 G
/F76 8.9664 Tf 258.75 632.885 Td [(T)80(able)-250(6.)-250(Comparison)-250(of)-250(captioning)-250(models)-250(on)-250(Landmarks)-250(dataset.)]TJ
ET
1 0 0 1 352.899 683.777 cm
q
.95 0 0 .95 0 0 cm
q
1 0 0 1 2.491 20.822 cm
[]0 d 0 J 0.398 w 0 0 m 153.484 0 l S
Q
1 0 0 1 -352.899 -683.777 cm
BT
/F76 9.9626 Tf 361.368 696.031 Td [(Captioning)-250(Model)-1519(FID)-1756(KID)]TJ
ET
q
1 0 0 1 355.39 692.245 cm
[]0 d 0 J 0.398 w 0 0 m 153.484 0 l S
Q
BT
/F76 9.9626 Tf 378.623 683.678 Td [(OF)74(A)-250([25])]TJ/F75 9.9626 Tf 67.486 0 Td [(13.48)-1200(0.002)]TJ/F76 9.9626 Tf -74.5 -11.956 Td [(ClipCap)-250([12])-2228(16.35)-1200(0.003)]TJ
ET
q
1 0 0 1 355.39 667.937 cm
[]0 d 0 J 0.398 w 0 0 m 153.484 0 l S
Q
1 0 0 1 352.899 683.777 cm
Q
0 g 0 G
0 g 0 G
1 0 0 1 -32.224 -198.59 cm
q
.35664 0 0 .35664 0 0 cm
q
596.16 0 0 475.2 0 0 cm
/Im7 Do
Q
Q
1 0 0 1 -320.675 -485.187 cm
BT
/F76 8.9664 Tf 308.862 474.229 Td [(Figure)-313(7.)-314(F)15(ailure)-313(cases;)-345(The)-313(w)10(ord)-314<94626c7572727994>-313(in)-313(captions)-313(generated)]TJ 0 -10.959 Td [(blurry)-250(images.)-310(Red)-250(box)15(es)-250(indicate)-250(generated)-250(re)15(gions.)]TJ
0 g 0 G
/F75 10.9589 Tf 0 -30.915 Td [(4.7.)-250(Limitations)]TJ/F76 9.9626 Tf 11.955 -18.893 Td [(Our)-223(method)-223(f)10(ails)-222(in)-223(certain)-223(cases.)-301(As)-222(sho)25(wn)-223(in)-223(Fig.)-301(7)-223(the)]TJ -11.955 -11.955 Td [(image)-388(captioning)-388(model)-389(generates)-388(a)-388(caption)-388(containing)-389(a)]TJ 0 -11.955 Td [(w)10(ord)-208<93626c7572727994>-207(for)-208(blurry)-208(images.)-296(As)-207(a)-208(result,)-216(the)-208(image)-208(gen-)]TJ 0 -11.955 Td [(eration)-271(model)-270(generates)-271(e)15(xaggeratedly)-270(blurry)-271(images)-271(when)]TJ 0 -11.955 Td [(gi)25(v)15(en)-279(captions)-279(with)-279(a)-279(w)10(ord)-279<93626c75727279942e>-397(Another)-279(limitation)-279(of)]TJ 0 -11.956 Td [(our)-272(CEP)-272(module)-272(emplo)10(ying)-272(GLIDE)-272([13])-272(is)-272(that)-272(the)-272(resolu-)]TJ 0 -11.955 Td [(tion)-250(of)-250(the)-250(output)-250(at)-250(each)-250(step)-250(is)-250(limited)-250(to)]TJ/F20 9.9626 Tf 170.468 0 Td [(256)]TJ/F26 9.9626 Tf 17.158 0 Td [<02>]TJ/F20 9.9626 Tf 9.963 0 Td [(256)]TJ/F76 9.9626 Tf 14.944 0 Td [(.)]TJ/F75 11.9552 Tf -212.533 -26.79 Td [(5.)-250(Conclusion)]TJ/F76 9.9626 Tf 11.955 -19.889 Td [(In)-301(this)-300(paper)40(,)-313(we)-301(propose)-300(a)-301(no)15(v)15(el)-300(modality)-301(of)-301(hints,)-313(the)]TJ -11.955 -11.955 Td [(natural)-353(language,)-378(and)-353(incorporate)-353(it)-352(to)-353(image)-353(outpainting,)]TJ 0 -11.956 Td [(and)-395(e)25(v)15(en)-395(to)-394(wide-range)-395(image)-395(blending)-395(tasks.)-744(Since)-395(the)]TJ 0 -11.955 Td [(proposed)-276(captioning-based)-277(Extensi)25(v)15(e)-276(P)15(ainting)-276(\(CEP\))-277(mod-)]TJ 0 -11.955 Td [(ule)-295(can)-295(adopt)-295(an)15(y)-294(image)-295(captioning)-295(and)-295(te)15(xt-guided)-295(image)]TJ 0 -11.955 Td [(generation)-365(algori)1(thms)-365(for)-365(each)-364(hint)-365(generation)-364(and)-365(image)]TJ 0 -11.955 Td [(generation)-207(netw)10(ork,)-216(we)-207(tak)10(e)-208(adv)25(antage)-207(of)-207(the)-208(well-designed)]TJ 0 -11.955 Td [(pretrained)-259(models.)-339(F)15(or)-259(both)-260(tasks,)-262(our)-259(module)-260(outperforms)]TJ 0 -11.956 Td [(the)-230(baseline)-231(models)-230(most)-231(of)-230(the)-231(times)1(,)-235(and)-230(generates)-231(photo-)]TJ 0 -11.955 Td [(realistic)-229(images)-228(e)25(v)15(en)-229(for)-229(semantically)-228(complicated)-229(images.)]TJ/F75 11.9552 Tf 0 -26.79 Td [(6.)-250(Ackno)10(wledgement)]TJ/F76 9.9626 Tf 11.955 -19.889 Td [(W)80(e)-622(ackno)25(wledge)-623(fundings)-622(from)-623(Basic)-622(Science)-623(Re-)]TJ -11.955 -11.955 Td [(search)-493(Program)-492(through)-493(the)-492(National)-493(Research)-493(F)15(ounda-)]TJ 0 -11.956 Td [(tion)-330(of)-331(K)35(orea)-330(\(NRF\))-331(funded)-330(by)-330(the)-331(Ministry)-330(of)-331(Education)]TJ 0 -11.955 Td [(\(2021R1I1A1A01051225\),)-231(and)-226(from)-226(the)-226(National)-226(Research)]TJ 0 -11.955 Td [(F)15(oundation)-227(of)-226(K)35(orea)-227(\(NRF\))-227(grant)-227(funded)-226(by)-227(the)-227(K)35(orea)-227(go)15(v-)]TJ 0 -11.955 Td [(ernment)-250(\(MSIT\))-250(\(No.)-310(2021R1A2C1004208\).)]TJ
0 g 0 G
0 g 0 G
ET
Q
Q
q
1 0 0 1 0 0 cm
/Z1GyqzNKlwlUjhEm0vHqEA Do
Q
endstream
endobj
32 0 obj
<< /Font << /F20 55 0 R /F26 58 0 R /F75 40 0 R /F76 41 0 R >> /ProcSet [ /PDF /Text /ImageC ] /XObject << /Im6 66 0 R /Im7 67 0 R /Z1GyqzNKlwlUjhEm0vHqEA 68 0 R >> >>
endobj
33 0 obj
<< /Length 11282 >>
stream
q
q
0 g 0 G
0 g 0 G
BT
/F75 11.9552 Tf 50.112 710.037 Td [(Refer)18(ences)]TJ
0 g 0 G
/F76 9.9626 Tf 4.981 -19.544 Td [([1])]TJ
0 g 0 G
[-500(Guilin)-287(Liu,)-297(Fitsum)-287(A)-287(Reda,)-297(K)25(e)25(vin)-287(J)-287(Shih,)-297(T)35(ing-Chun)]TJ 16.598 -11.956 Td [(W)80(ang,)-302(Andre)25(w)-292(T)80(ao,)-303(and)-292(Bryan)-292(Catanzaro.)-494(Image)-292(in-)]TJ 0 -11.955 Td [(painting)-248(for)-247(irre)15(gular)-248(holes)-247(using)-248(partial)-248(con)40(v)20(olutions.)]TJ 0 -11.955 Td [(In)]TJ/F80 9.9626 Tf 11.875 0 Td [(Pr)45(oceedings)-359(of)-359(the)-359(Eur)45(opean)-359(confer)37(ence)-359(on)-359(com-)]TJ -11.875 -11.955 Td [(puter)-250(vision)-250(\(ECCV\))]TJ/F76 9.9626 Tf 81.912 0 Td [(,)-250(pages)-250(85\226100,)-250(2018.)]TJ
0 g 0 G
-98.51 -22.388 Td [([2])]TJ
0 g 0 G
[-500(Deepak)-452(P)15(athak,)-503(Philipp)-452(Krahenb)20(uhl,)-503(Jef)25(f)-452(Donahue,)]TJ 16.598 -11.956 Td [(T)35(re)25(v)20(or)-220(Darrell,)-226(and)-221(Ale)15(x)15(ei)-220(A)-220(Efros.)-296(C)1(o)-1(nt)1(e)15(x)-1(t)-220(encoders:)]TJ 0 -11.955 Td [(Feature)-468(learning)-469(by)-468(inpainting.)-1059(In)]TJ/F80 9.9626 Tf 152.899 0 Td [(Pr)45(oceedings)-468(of)]TJ -152.899 -11.955 Td [(the)-241(IEEE)-240(Confer)37(ence)-241(on)-241(Computer)-240(V)74(ision)-241(and)-241(P)80(attern)]TJ 0 -11.955 Td [(Reco)10(gnition)]TJ/F76 9.9626 Tf 48.049 0 Td [(,)-250(pages)-250(2536\2262544,)-250(2016.)]TJ
0 g 0 G
-64.647 -22.388 Td [([3])]TJ
0 g 0 G
[-500(Jiahui)-194(Y)111(u,)-205(Zhe)-193(Lin,)-205(Jimei)-194(Y)100(ang,)-205(Xiaohui)-193(Shen,)-205(Xin)-194(Lu,)]TJ 16.598 -11.956 Td [(and)-366(Thomas)-366(S)-366(Huang.)-732(Generati)25(v)15(e)-367(image)-366(inpainting)]TJ 0 -11.955 Td [(with)-242(conte)15(xtual)-242(attention.)]TJ/F80 9.9626 Tf 104.674 0 Td [(In)-242(The)-243(IEEE)-242(Confer)37(ence)-242(on)]TJ -104.674 -11.955 Td [(Computer)-250(V)74(ision)-250(and)-250(P)80(attern)-250(Reco)10(gnition)]TJ/F76 9.9626 Tf 166.624 0 Td [(,)-250(2018.)]TJ
0 g 0 G
-183.222 -22.388 Td [([4])]TJ
0 g 0 G
[-500(Piotr)-348(T)70(eterw)10(ak,)-372(Aaron)-348(Sarna,)-372(Dilip)-348(Krishnan,)-372(Aaron)]TJ 16.598 -11.955 Td [(Maschinot,)-418(Da)20(vid)-384(Belanger)40(,)-417(Ce)-384(Liu,)-418(and)-384(W)40(illiam)-384(T)]TJ 0 -11.956 Td [(Freeman.)-271(Boundless:)-290(Generati)25(v)15(e)-209(adv)15(ersarial)-209(netw)10(orks)]TJ 0 -11.955 Td [(for)-236(image)-237(e)15(xtension.)-331(In)]TJ/F80 9.9626 Tf 95.155 0 Td [(Pr)45(oceedings)-236(of)-237(the)-236(IEEE/CVF)]TJ -95.155 -11.955 Td [(International)-333(Confer)37(ence)-334(on)-333(Computer)-333(V)74(ision)]TJ/F76 9.9626 Tf 185.966 0 Td [(,)-354(pages)]TJ -185.966 -11.955 Td [(10521\22610530,)-250(2019.)]TJ
0 g 0 G
-16.598 -22.388 Td [([5])]TJ
0 g 0 G
[-500(Y)55(i)-453(W)80(ang,)-504(Xin)-453(T)80(ao,)-503(Xiao)10(yong)-453(Shen,)-504(and)-453(Jiaya)-453(Jia.)]TJ 16.598 -11.956 Td [(W)40(ide-conte)15(xt)-343(semantic)-342(image)-343(e)15(xtrapolation.)-657(In)]TJ/F80 9.9626 Tf 196.858 0 Td [(Pr)45(o-)]TJ -196.858 -11.955 Td [(ceedings)-357(of)-357(the)-357(IEEE/CVF)-357(Confer)37(ence)-358(on)-357(Computer)]TJ 0 -11.955 Td [(V)74(ision)-497(and)-497(P)80(attern)-496(Reco)10(gnition)]TJ/F76 9.9626 Tf 131.659 0 Td [(,)-559(pages)-496(1399\2261408,)]TJ -131.659 -11.955 Td [(2019.)]TJ
0 g 0 G
-16.598 -22.388 Td [([6])]TJ
0 g 0 G
[-500(Zongxin)-426(Y)100(ang,)-469(Jian)-426(Dong,)-470(Ping)-426(Liu,)-469(Y)55(i)-426(Y)100(ang,)-470(and)]TJ 16.598 -11.956 Td [(Shuicheng)-489(Y)100(an.)-1124(V)111(ery)-489(long)-489(natural)-489(scenery)-489(image)]TJ 0 -11.955 Td [(prediction)-529(by)-529(outpainting.)-1254(In)]TJ/F80 9.9626 Tf 134.846 0 Td [(Pr)45(oceedings)-529(of)-529(the)]TJ -134.846 -11.955 Td [(IEEE/CVF)-243(International)-243(Confer)37(ence)-243(on)-244(Computer)-243(V)74(i-)]TJ 0 -11.955 Td [(sion)]TJ/F76 9.9626 Tf 16.608 0 Td [(,)-250(pages)-250(10561\22610570,)-250(2019.)]TJ
0 g 0 G
-33.206 -22.388 Td [([7])]TJ
0 g 0 G
[-500(Chia-Ni)-538(Lu,)-609(Y)100(a-Chu)-537(Chang,)-610(and)-537(W)80(ei-Chen)-538(Chiu.)]TJ 16.598 -11.956 Td [(Bridging)-283(the)-283(visual)-282(g)5(ap:)-376(W)40(ide-range)-283(image)-283(blending.)]TJ 0 -11.955 Td [(In)]TJ/F80 9.9626 Tf 10.623 0 Td [(Pr)45(oceedings)-233(of)-233(the)-234(IEEE/CVF)-233(Confer)37(ence)-233(on)-233(Com-)]TJ -10.623 -11.955 Td [(puter)-239(V)74(ision)-238(and)-239(P)80(attern)-239(Reco)10(gnition)]TJ/F76 9.9626 Tf 147.356 0 Td [(,)-241(pages)-239(843\226851,)]TJ -147.356 -11.955 Td [(2021.)]TJ
0 g 0 G
-16.598 -22.388 Td [([8])]TJ
0 g 0 G
[-500(K)25(yunghun)-685(Kim,)-793(Y)100(eohun)-685(Y)111(un,)-794(K)25(eon-W)80(oo)-685(Kang,)]TJ 16.598 -11.956 Td [(K)25(yeongbo)-528(K)35(ong,)-597(Siyeong)-527(Lee,)-597(and)-528(Suk-Ju)-528(Kang.)]TJ 0 -11.955 Td [(P)15(ainting)-400(outside)-399(as)-400(inside:)-610(Edge)-399(guided)-400(image)-400(out-)]TJ 0 -11.955 Td [(painting)-222(via)-221(bidirectional)-222(rearrangement)-221(with)-222(progres-)]TJ 0 -11.955 Td [(si)25(v)15(e)-343(step)-344(learning.)-658(In)]TJ/F80 9.9626 Tf 91.962 0 Td [(Pr)45(oceedings)-343(of)-344(the)-343(IEEE/CVF)]TJ -91.962 -11.955 Td [(W)55(inter)-388(Confer)37(ence)-387(on)-388(Applications)-387(of)-388(Computer)-388(V)74(i-)]TJ 0 -11.956 Td [(sion)]TJ/F76 9.9626 Tf 16.608 0 Td [(,)-250(pages)-250(2122\2262130,)-250(2021.)]TJ
0 g 0 G
-33.206 -22.388 Td [([9])]TJ
0 g 0 G
[-500(Naofumi)-390(Akimoto,)-424(Daiki)-390(Ito,)-424(and)-390(Y)110(oshimitsu)-390(Aoki.)]TJ 16.598 -11.955 Td [(Scenery)-409(image)-409(e)15(xtension)-409(via)-409(inpainting)-409(with)-409(a)-409(mir)20(-)]TJ 0 -11.955 Td [(rored)-250(input.)]TJ/F80 9.9626 Tf 50.072 0 Td [(IEEE)-250(Access)]TJ/F76 9.9626 Tf 51.177 0 Td [(,)-250(9:59286\22659300,)-250(2021.)]TJ
0 g 0 G
0 g 0 G
0 g 0 G
135.922 629.037 Td [([10])]TJ
0 g 0 G
[-500(Daeh)5(yeon)-433(K)35(ong,)-480(K)25(yeongbo)-433(K)35(ong,)-479(K)25(yunghun)-434(Kim,)]TJ 21.579 -11.955 Td [(Sung-Jun)-231(Min,)-235(and)-231(Suk-Ju)-231(Kang.)-320(Image-adapti)25(v)15(e)-231(hint)]TJ 0 -11.955 Td [(generation)-321(via)-320(vision)-321(transformer)-321(for)-320(outpainting.)-587(In)]TJ/F80 9.9626 Tf 0 -11.955 Td [(Pr)45(oceedings)-337(of)-337(the)-336(IEEE/CVF)-337(W)55(inter)-337(Confer)37(ence)-337(on)]TJ 0 -11.956 Td [(Applications)-384(of)-384(Computer)-385(V)74(ision)]TJ/F76 9.9626 Tf 134.183 0 Td [(,)-418(pages)-384(3572\2263581,)]TJ -134.183 -11.955 Td [(2022.)]TJ
0 g 0 G
-21.579 -23.337 Td [([11])]TJ
0 g 0 G
[-500(Peng)-327(W)80(ang,)-345(An)-327(Y)100(ang,)-345(Rui)-327(Men,)-346(Jun)15(yang)-326(Lin,)-346(Shuai)]TJ 21.579 -11.955 Td [(Bai,)-453(Zhikang)-412(Li,)-453(Jianxin)-412(Ma,)-453(Chang)-413(Zhou,)-453(Jingren)]TJ 0 -11.955 Td [(Zhou,)-527(and)-472(Hongxia)-472(Y)100(ang.)-1070(Unifying)-472(architectures,)]TJ 0 -11.955 Td [(tasks,)-421(and)-387(modalities)-387(through)-387(a)-387(simple)-387(sequence-to-)]TJ 0 -11.955 Td [(sequence)-335(learning)-334(frame)25(w)10(ork.)]TJ/F80 9.9626 Tf 127.961 0 Td [(International)-335(Confer)20(-)]TJ -127.961 -11.956 Td [(ence)-250(on)-250(Mac)15(hine)-250(Learning)]TJ/F76 9.9626 Tf 106.928 0 Td [(,)-250(2022.)]TJ
0 g 0 G
-128.507 -23.337 Td [([12])]TJ
0 g 0 G
[-500(Ron)-190(Mokady)65(,)-203(Amir)-190(Hertz,)-202(and)-190(Amit)-191(H)-190(Bermano.)-231(Clip-)]TJ 21.579 -11.955 Td [(cap:)-381(Clip)-285<7072650278>-285(for)-286(image)-285(captioning.)]TJ/F80 9.9626 Tf 157.403 0 Td [(arXiv)-285(pr)37(eprint)]TJ -157.403 -11.955 Td [(arXiv:2111.09734)]TJ/F76 9.9626 Tf 72.777 0 Td [(,)-250(2021.)]TJ
0 g 0 G
-94.356 -23.337 Td [([13])]TJ
0 g 0 G
[-500(Thomas)-477(A)-477(Halgren,)-534(Robert)-477(B)-477(Murph)5(y)65(,)-534(Richard)-478(A)]TJ 21.579 -11.955 Td [(Friesner)40(,)-250(He)15(ge)-249(S)-250(Beard,)-249(Leah)-250(L)-249(Frye,)-250(W)-249(Thomas)-250(Pol-)]TJ 0 -11.955 Td [(lard,)-428(and)-392(Jay)-392(L)-392(Banks.)-814(Glide:)-594(a)-393(ne)25(w)-392(approach)-392(for)]TJ 0 -11.956 Td [(rapid,)-511(accurate)-458(docking)-458(and)-459(scoring.)-458(2.)-459(enrichment)]TJ 0 -11.955 Td [(f)10(actors)-386(in)-386(database)-386(screening.)]TJ/F80 9.9626 Tf 129.191 0 Td [(J)25(ournal)-386(of)-386(medicinal)]TJ -129.191 -11.955 Td [(c)15(hemistry)]TJ/F76 9.9626 Tf 38.585 0 Td [(,)-250(47\(7\):1750\2261759,)-250(2004.)]TJ
0 g 0 G
-60.164 -23.337 Td [([14])]TJ
0 g 0 G
[-500(Omri)-470(A)74(vrahami,)-525(Dani)-470(Lischinski,)-525(and)-470(Ohad)-470(Fried.)]TJ 21.579 -11.955 Td [(Blended)-193(dif)25(fusion)-193(for)-194(te)15(xt-dri)25(v)15(en)-193(editing)-193(of)-193(natural)-194(im-)]TJ 0 -11.955 Td [(ages.)-835(In)]TJ/F80 9.9626 Tf 40.785 0 Td [(Pr)45(oceedings)-399(of)-398(the)-399(IEEE/CVF)-398(Confer)37(ence)]TJ -40.785 -11.955 Td [(on)-377(Computer)-378(V)74(ision)-377(and)-377(P)80(attern)-378(Reco)10(gnition)]TJ/F76 9.9626 Tf 185.419 0 Td [(,)-409(pages)]TJ -185.419 -11.956 Td [(18208\22618218,)-250(2022.)]TJ
0 g 0 G
-21.579 -23.337 Td [([15])]TJ
0 g 0 G
[-500(Christoph)-231(Schuhmann,)-234(Richard)-231(V)111(encu,)-234(Romain)-231(Beau-)]TJ 21.579 -11.955 Td [(mont,)-382(Robert)-355(Kaczmarczyk,)-382(Clayton)-355(Mullis,)-382(Aarush)]TJ 0 -11.955 Td [(Katta,)-547(Theo)-488(Coombes,)-547(Jenia)-488(Jitse)25(v)65(,)-547(and)-488(Aran)-488(K)35(o-)]TJ 0 -11.955 Td [(matsuzaki.)-1528(Lai)1(on-400m:)-1040(Open)-615(dataset)-615(of)-615(clip-)]TJ 0 -11.955 Td [<026c7465726564>-347(400)-347(million)-347(image-te)15(xt)-347(pairs.)]TJ/F80 9.9626 Tf 156.788 0 Td [(arXiv)-347(pr)37(eprint)]TJ -156.788 -11.955 Td [(arXiv:2111.02114)]TJ/F76 9.9626 Tf 72.777 0 Td [(,)-250(2021.)]TJ
0 g 0 G
-94.356 -23.338 Td [([16])]TJ
0 g 0 G
[-500(Coloma)-1012(Ballester)40(,)-1202(Marcelo)-1012(Bertalmio,)-1202(V)60(icent)]TJ 21.579 -11.955 Td [(Caselles,)-956(Guillermo)-815(Sapiro,)-956(and)-815(Joan)-815(V)111(erdera.)]TJ 0 -11.955 Td [(Filling-in)-414(by)-413(joint)-414(interpolation)-414(of)-413(v)15(ector)-414<02656c6473>-414(and)]TJ 0 -11.955 Td [(gray)-318(le)25(v)15(els.)]TJ/F80 9.9626 Tf 51.972 0 Td [(IEEE)-318(tr)15(ansactions)-318(on)-319(ima)10(g)10(e)-318(pr)45(ocessing)]TJ/F76 9.9626 Tf 160.208 0 Td [(,)]TJ -212.18 -11.955 Td [(10\(8\):1200\2261211,)-250(2001.)]TJ
0 g 0 G
-21.579 -23.337 Td [([17])]TJ
0 g 0 G
[-500(Marcelo)-892(Bertalmio,)-1052(Guillermo)-892(Sapiro,)-1052(V)60(incent)]TJ 21.579 -11.955 Td [(Caselles,)-279(and)-273(Coloma)-273(Ballester)55(.)-434(Image)-273(inpainting.)-434(In)]TJ/F80 9.9626 Tf 0 -11.956 Td [(Pr)45(oceedings)-352(of)-352(the)-353(27th)-352(annual)-352(confer)37(ence)-352(on)-353(Com-)]TJ 0 -11.955 Td [(puter)-233(gr)15(aphics)-233(and)-233(inter)15(active)-233(te)1(c)14(hniques)]TJ/F76 9.9626 Tf 164.896 0 Td [(,)-236(pages)-233<34313796>]TJ -164.896 -11.955 Td [(424,)-250(2000.)]TJ
0 g 0 G
-21.579 -23.337 Td [([18])]TJ
0 g 0 G
[-500(Ale)15(x)15(ei)-223(A)-223(Efros)-223(and)-223(Thomas)-223(K)-222(Leung.)-302(T)70(e)15(xture)-223(synthe-)]TJ 21.579 -11.955 Td [(sis)-228(by)-228(non-parametric)-227(sampling.)-312(In)]TJ/F80 9.9626 Tf 140.852 0 Td [(Pr)45(oceedings)-228(of)-228(the)]TJ -140.852 -11.955 Td [(se)15(venth)-410(IEEE)-409(international)-410(confer)37(ence)-409(on)-410(computer)]TJ 0 -11.955 Td [(vision)]TJ/F76 9.9626 Tf 23.8 0 Td [(,)-250(v)20(olume)-250(2,)-250(pages)-250(1033\2261038.)-250(IEEE,)-250(1999.)]TJ
0 g 0 G
0 g 0 G
ET
Q
Q
q
1 0 0 1 0 0 cm
/iGtSExez6OAdpEweR0ZH2w Do
Q
endstream
endobj
34 0 obj
<< /Font << /F75 40 0 R /F76 41 0 R /F80 43 0 R >> /ProcSet [ /PDF /Text ] /XObject << /iGtSExez6OAdpEweR0ZH2w 69 0 R >> >>
endobj
35 0 obj
<< /Length 12084 >>
stream
q
q
0 g 0 G
0 g 0 G
0 g 0 G
BT
/F76 9.9626 Tf 50.112 710.037 Td [([19])]TJ
0 g 0 G
[-500(Josef)-383(Si)25(vic,)-417(Biliana)-383(Kane)25(v)25(a,)-417(Antonio)-383(T)80(orralba,)-417(Shai)]TJ 21.579 -11.955 Td [(A)74(vidan,)-228(and)-223(W)40(illiam)-223(T)-223(Freeman.)-302(Creating)-223(and)-223(e)15(xplor)20(-)]TJ 0 -11.955 Td [(ing)-249(a)-248(lar)18(ge)-249(photorealistic)-249(virtual)-248(space.)-357(In)]TJ/F80 9.9626 Tf 168.478 0 Td [(IEEE)-249(Com-)]TJ -168.478 -11.955 Td [(puter)-193(Society)-193(Confer)37(ence)-193(on)-193(Computer)-193(V)74(ision)-194(and)-193(P)80(at-)]TJ 0 -11.955 Td [(tern)-250(Reco)10(gnition)-250(W)92(orkshops)]TJ/F76 9.9626 Tf 112.437 0 Td [(,)-250(pages)-250<3196382e>-250(IEEE,)-250(2008.)]TJ
0 g 0 G
-134.016 -19.416 Td [([20])]TJ
0 g 0 G
[-500(Y)55(inda)-314(Zhang,)-329(Jianxiong)-314(Xiao,)-330(James)-313(Hays,)-330(and)-314(Ping)]TJ 21.579 -11.955 Td [(T)80(an.)-720(Framebreak:)-535(Dramati)1(c)-363(image)-362(e)15(xtrapolation)-363(by)]TJ 0 -11.955 Td [(guided)-309(shift-maps.)-548(In)]TJ/F80 9.9626 Tf 91.583 0 Td [(Pr)45(oceedings)-309(of)-309(t)1(h)-1(e)-308(IEEE)-309(Con-)]TJ -91.583 -11.956 Td [(fer)37(ence)-278(on)-278(Computer)-277(V)74(ision)-278(and)-278(P)80(attern)-278(Reco)10(gnition)]TJ/F76 9.9626 Tf 212.18 0 Td [(,)]TJ -212.18 -11.955 Td [(pages)-250(1171\2261178,)-250(2013.)]TJ
0 g 0 G
-21.579 -19.415 Td [([21])]TJ
0 g 0 G
[-500(Justin)-563(Johnson,)-642(Andrej)-563(Karpath)5(y)65(,)-641(and)-564(Li)-563(Fei-Fei.)]TJ 21.579 -11.956 Td [(Densecap:)-490(Ful)1(ly)-340(con)40(v)20(olutional)-340(localization)-340(netw)10(orks)]TJ 0 -11.955 Td [(for)-203(dense)-203(captioning.)-258(In)]TJ/F80 9.9626 Tf 95.795 0 Td [(Pr)45(oceedings)-203(of)-203(the)-203(IEEE)-203(Con-)]TJ -95.795 -11.955 Td [(fer)37(ence)-278(on)-278(Computer)-277(V)74(ision)-278(and)-278(P)80(attern)-278(Reco)10(gnition)]TJ/F76 9.9626 Tf 212.18 0 Td [(,)]TJ -212.18 -11.955 Td [(pages)-250(4565\2264574,)-250(2016.)]TJ
0 g 0 G
-21.579 -19.416 Td [([22])]TJ
0 g 0 G
[-500(Andrej)-192(Karpath)5(y)-192(and)-193(Li)-192(Fei-Fei.)-235(Deep)-192(visual-semantic)]TJ 21.579 -11.955 Td [(alignments)-252(for)-251(generating)-252(image)-252(descriptions.)-365(In)]TJ/F80 9.9626 Tf 196.858 0 Td [(Pr)45(o-)]TJ -196.858 -11.955 Td [(ceedings)-261(of)-261(the)-261(IEEE)-261(Confer)37(ence)-261(on)-262(Computer)-261(V)74(ision)]TJ 0 -11.955 Td [(and)-250(P)80(attern)-250(Reco)10(gnition)]TJ/F76 9.9626 Tf 97.065 0 Td [(,)-250(pages)-250(3128\2263137,)-250(2015.)]TJ
0 g 0 G
-118.644 -19.416 Td [([23])]TJ
0 g 0 G
[-500(An-An)-442(Liu,)-491(Y)55(ingchen)-443(Zhai,)-490(Ning)-443(Xu,)-490(W)80(eizhi)-443(Nie,)]TJ 21.579 -11.955 Td [(W)80(enhui)-297(Li,)-308(and)-296(Y)110(ongdong)-297(Zhang.)-509(Re)15(gion-a)15(w)10(are)-297(im-)]TJ 0 -11.955 Td [(age)-316(captioning)-315(via)-316(interaction)-316(learning.)]TJ/F80 9.9626 Tf 164.074 0 Td [(IEEE)-316(T)55(r)15(ans-)]TJ -164.074 -11.955 Td [(actions)-226(on)-226(Cir)37(cuits)-226(and)-226(Systems)-226(for)-226(V)74(ideo)-226(T)92(ec)15(hnolo)10(gy)]TJ/F76 9.9626 Tf 212.18 0 Td [(,)]TJ -212.18 -11.956 Td [(2021.)]TJ
0 g 0 G
-21.579 -19.415 Td [([24])]TJ
0 g 0 G
[-500(Ste)25(v)15(en)-232(J)-231(Rennie,)-235(Etienne)-232(Marcheret,)-235(Y)110(oussef)-232(Mroueh,)]TJ 21.579 -11.955 Td [(Jerret)-224(Ross,)-230(and)-224(V)111(aibha)20(v)25(a)-225(Goel.)-304(Self-critical)-225(sequence)]TJ 0 -11.956 Td [(training)-327(for)-327(image)-327(captioning.)-606(In)]TJ/F80 9.9626 Tf 138.877 0 Td [(Pr)45(oceedings)-327(of)-327(the)]TJ -138.877 -11.955 Td [(IEEE)-484(Confer)37(ence)-485(on)-484(Computer)-485(V)74(ision)-484(and)-485(P)80(attern)]TJ 0 -11.955 Td [(Reco)10(gnition)]TJ/F76 9.9626 Tf 48.049 0 Td [(,)-250(pages)-250(7008\2267024,)-250(2017.)]TJ
0 g 0 G
-69.628 -19.416 Td [([25])]TJ
0 g 0 G
[-500(Luo)25(wei)-374(Zhou,)-406(Hamid)-374(P)15(alangi,)-406(Lei)-374(Zhang,)-406(Houdong)]TJ 21.579 -11.955 Td [(Hu,)-357(Jason)-335(Corso,)-357(and)-335(Jianfeng)-336(Gao.)-633<556e69026564>-336(vision-)]TJ 0 -11.955 Td [(language)-374(pre-training)-375(for)-374(image)-375(captioning)-374(and)-375(vqa.)]TJ 0 -11.955 Td [(In)]TJ/F80 9.9626 Tf 11.808 0 Td [(Pr)45(oceedings)-352(of)-353(the)-352(AAAI)-352(Confer)37(ence)-352(on)-353(Ar)1<7469026369616c>]TJ -11.808 -11.955 Td [(Intellig)10(ence)]TJ/F76 9.9626 Tf 46.933 0 Td [(,)-250(v)20(olume)-250(34,)-250(pages)-250(13041\22613049,)-250(2020.)]TJ
0 g 0 G
-68.512 -19.416 Td [([26])]TJ
0 g 0 G
[-500(Xiujun)-348(Li,)-372(Xi)-348(Y)55(in,)-372(Chun)15(yuan)-348(Li,)-372(Pengchuan)-348(Zhang,)]TJ 21.579 -11.955 Td [(Xiao)25(wei)-310(Hu,)-324(Lei)-310(Zhang,)-324(Lijuan)-310(W)80(ang,)-324(Houdong)-310(Hu,)]TJ 0 -11.955 Td [(Li)-371(Dong,)-401(Furu)-371(W)80(ei,)-402(et)-371(al.)-747(Oscar:)-552(Object-semantics)]TJ 0 -11.956 Td [(aligned)-273(pre-training)-274(for)-273(vision-language)-273(tasks.)-435(In)]TJ/F80 9.9626 Tf 200.285 0 Td [(Eu-)]TJ -200.285 -11.955 Td [(r)45(opean)-358(Confer)37(ence)-359(on)-358(Computer)-359(V)74(ision)]TJ/F76 9.9626 Tf 162.158 0 Td [(,)-386(pages)-358<31323196>]TJ -162.158 -11.955 Td [(137.)-250(Springer)40(,)-250(2020.)]TJ
0 g 0 G
-21.579 -19.416 Td [([27])]TJ
0 g 0 G
[-500(Bo)25(wen)-434(Li,)-479(Xiaojuan)-434(Qi,)-479(Thomas)-434(Lukasie)25(wicz,)-480(and)]TJ 21.579 -11.955 Td [(Philip)-295(HS)-294(T)80(orr)55(.)-503(Manig)5(an:)-399(T)70(e)15(xt-guided)-294(image)-295(manip-)]TJ 0 -11.955 Td [(ulation.)-445(In)]TJ/F80 9.9626 Tf 45.649 0 Td [(Pr)45(oceedings)-276(of)-277(the)-276(IEEE/CVF)-277(Confer)37(ence)]TJ -45.649 -11.955 Td [(on)-377(Computer)-378(V)74(ision)-377(and)-377(P)80(attern)-378(Reco)10(gnition)]TJ/F76 9.9626 Tf 185.419 0 Td [(,)-409(pages)]TJ -185.419 -11.955 Td [(7880\2267889,)-250(2020.)]TJ
0 g 0 G
-21.579 -19.416 Td [([28])]TJ
0 g 0 G
[-500(Lisai)-264(Zhang,)-267(Qingcai)-263(Chen,)-267(Baotian)-264(Hu,)-267(and)-264(Shuoran)]TJ 21.579 -11.955 Td [(Jiang.)-514(T)70(e)15(xt-guided)-298(neural)-298(image)-298(inpainting.)-515(In)]TJ/F80 9.9626 Tf 193.939 0 Td [(A)30(CM)]TJ -193.939 -11.955 Td [(International)-250(Confer)37(ence)-250(on)-250(Multimedia)]TJ/F76 9.9626 Tf 161.513 0 Td [(,)-250(2020.)]TJ
0 g 0 G
0 g 0 G
0 g 0 G
75.658 629.037 Td [([29])]TJ
0 g 0 G
[-500(Li)-722(Zhang,)-839(Flood)-722(Sung,)-840(Feng)-721(Liu,)-840(T)80(ao)-722(Xiang,)]TJ 21.579 -11.955 Td [(Shaog)5(ang)-587(Gong,)-672(Y)110(ongxin)-587(Y)100(ang,)-672(and)-587(T)35(imoth)5(y)-588(M)]TJ 0 -11.955 Td [(Hospedales.)-492(Actor)20(-critic)-291(sequence)-291(training)-291(for)-292(image)]TJ 0 -11.955 Td [(captioning.)]TJ/F80 9.9626 Tf 48.139 0 Td [(arXiv)-250(pr)37(eprint)-250(arXiv:1706.09601)]TJ/F76 9.9626 Tf 132.183 0 Td [(,)-250(2017.)]TJ
0 g 0 G
-201.901 -20.349 Td [([30])]TJ
0 g 0 G
[-500(Junlong)-288(Gao,)-297(Shiqi)-288(W)80(ang,)-297(Shanshe)-288(W)80(ang,)-297(Siwei)-288(Ma,)]TJ 21.579 -11.955 Td [(and)-319(W)80(en)-319(Gao.)-580(Self-critical)-319(n-step)-319(training)-319(for)-319(image)]TJ 0 -11.955 Td [(captioning.)-885(In)]TJ/F80 9.9626 Tf 65.797 0 Td [(Pr)45(oceedings)-414(of)-414(the)-414(IEEE/CVF)-414(Con-)]TJ -65.797 -11.955 Td [(fer)37(ence)-278(on)-278(Computer)-277(V)74(ision)-278(and)-278(P)80(attern)-278(Reco)10(gnition)]TJ/F76 9.9626 Tf 212.18 0 Td [(,)]TJ -212.18 -11.955 Td [(pages)-250(6300\2266308,)-250(2019.)]TJ
0 g 0 G
-21.579 -20.349 Td [([31])]TJ
0 g 0 G
[-500(Ronald)-640(J)-640(W)39(i)1(lliams.)-1610(Simple)-640(statistical)-641(gradient-)]TJ 21.579 -11.955 Td [(follo)25(wing)-324(algorithms)-325(for)-324(connectionist)-325(reinforcement)]TJ 0 -11.955 Td [(learning.)]TJ/F80 9.9626 Tf 38.724 0 Td [(Mac)15(hine)-250(learning)]TJ/F76 9.9626 Tf 70.964 0 Td [(,)-250(8\(3\):229\226256,)-250(1992.)]TJ
0 g 0 G
-131.267 -20.348 Td [([32])]TJ
0 g 0 G
[-500(Peter)-210(Anderson,)-219(Xiaodong)-210(He,)-218(Chris)-210(Buehler)40(,)-219(Damien)]TJ 21.579 -11.955 Td [(T)70(ene)15(y)65(,)-244(Mark)-242(Johnson,)-244(Stephen)-242(Gould,)-244(and)-242(Lei)-243(Zhang.)]TJ 0 -11.955 Td [(Bottom-up)-262(and)-263(top-do)25(wn)-262(attention)-263(for)-262(image)-263(caption-)]TJ 0 -11.956 Td [(ing)-279(and)-278(visual)-279(question)-278(answering.)-452(In)]TJ/F80 9.9626 Tf 154.791 0 Td [(Pr)45(oceedings)-278(of)]TJ -154.791 -11.955 Td [(the)-241(IEEE)-240(Confer)37(ence)-241(on)-241(Computer)-240(V)74(ision)-241(and)-241(P)80(attern)]TJ 0 -11.955 Td [(Reco)10(gnition)]TJ/F76 9.9626 Tf 48.049 0 Td [(,)-250(pages)-250(6077\2266086,)-250(2018.)]TJ
0 g 0 G
-69.628 -20.348 Td [([33])]TJ
0 g 0 G
[-500(Alec)-239(Radford,)-242(Jong)-240(W)80(ook)-239(Kim,)-242(Chris)-239(Hallac)15(y)65(,)-242(Aditya)]TJ 21.579 -11.955 Td [(Ramesh,)-278(Gabriel)-272(Goh,)-277(Sandhini)-272(Ag)5(arw)10(al,)-278(Girish)-272(Sas-)]TJ 0 -11.956 Td [(try)65(,)-574(Amanda)-510(Ask)10(ell,)-574(P)15(amela)-509(Mishkin,)-574(Jack)-510(Clark,)]TJ 0 -11.955 Td [(et)-209(al.)-271(Learning)-209(transferable)-209(visual)-209(models)-209(from)-209(natural)]TJ 0 -11.955 Td [(language)-255(supervision.)-374(In)]TJ/F80 9.9626 Tf 101.495 0 Td [(International)-255(Confer)37(ence)-254(on)]TJ -101.495 -11.955 Td [(Mac)15(hine)-250(Learning)]TJ/F76 9.9626 Tf 73.733 0 Td [(,)-250(pages)-250(8748\2268763.)-250(PMLR,)-250(2021.)]TJ
0 g 0 G
-95.312 -20.348 Td [([34])]TJ
0 g 0 G
[-500(Aditya)-361(Ramesh,)-388(Mikhail)-360(P)15(a)20(vlo)14(v)65(,)-388(Gabriel)-360(Goh,)-389(Scott)]TJ 21.579 -11.955 Td [(Gray)65(,)-397(Chelsea)-368(V)129(oss,)-397(Alec)-367(Radford,)-397(Mark)-368(Chen,)-397(and)]TJ 0 -11.956 Td [(Ilya)-232(Sutsk)10(e)25(v)15(er)55(.)-321(Zero-shot)-232(te)15(xt-to-image)-232(generation.)-322(In)]TJ/F80 9.9626 Tf 0 -11.955 Td [(International)-202(Confer)37(ence)-202(on)-201(Mac)15(hine)-202(Learning)]TJ/F76 9.9626 Tf 187.389 0 Td [(,)-211(pages)]TJ -187.389 -11.955 Td [(8821\2268831.)-250(PMLR,)-250(2021.)]TJ
0 g 0 G
-21.579 -20.348 Td [([35])]TJ
0 g 0 G
[-500(Chao)-276(Jia,)-282(Y)55(infei)-275(Y)100(ang,)-282(Y)100(e)-276(Xia,)-282(Y)55(i-T)35(ing)-276(Chen,)-282(Zarana)]TJ 21.579 -11.955 Td [(P)15(arekh,)-248(Hieu)-247(Pham,)-248(Quoc)-247(Le,)-248(Y)111(un-Hsuan)-247(Sung,)-248(Zhen)]TJ 0 -11.956 Td [(Li,)-418(and)-385(T)80(om)-385(Duerig.)-791(Scaling)-385(up)-384(visual)-385(and)-385(vision-)]TJ 0 -11.955 Td [(language)-391(representation)-390(learning)-391(with)-390(noisy)-391(te)15(xt)-391(su-)]TJ 0 -11.955 Td [(pervision.)-826(In)]TJ/F80 9.9626 Tf 60.036 0 Td [(International)-396(Confer)37(ence)-395(on)-396(Mac)15(hine)]TJ -60.036 -11.955 Td [(Learning)]TJ/F76 9.9626 Tf 36.533 0 Td [(,)-250(pages)-250(4904\2264916.)-250(PMLR,)-250(2021.)]TJ
0 g 0 G
-58.112 -20.348 Td [([36])]TJ
0 g 0 G
[-500(Chitw)10(an)-554(Saharia,)-629(W)40(illiam)-554(Chan,)-630(Saurabh)-554(Sax)15(ena,)]TJ 21.579 -11.955 Td [(Lala)-517(Li,)-583(Jay)-516(Whang,)-583(Emily)-517(Denton,)-583(Se)15(yed)-517(Kam-)]TJ 0 -11.956 Td [(yar)-273(Se)15(yed)-273(Ghasemipour)40(,)-278(Burcu)-273(Karagol)-273(A)92(yan,)-279(S)-273(Sara)]TJ 0 -11.955 Td [(Mahda)20(vi,)-440(Rapha)-402(Gontijo)-401(Lopes,)-440(et)-402(al.)-846(Photorealis-)]TJ 0 -11.955 Td [(tic)-237(te)15(xt-to-image)-238(dif)25(fusion)-237(models)-237(with)-237(deep)-238(language)]TJ 0 -11.955 Td [(understanding.)]TJ/F80 9.9626 Tf 73.277 0 Td [(arXiv)-587(pr)37(eprint)-587(arXiv:2205.11487)]TJ/F76 9.9626 Tf 138.903 0 Td [(,)]TJ -212.18 -11.955 Td [(2022.)]TJ
0 g 0 G
-21.579 -20.349 Td [([37])]TJ
0 g 0 G
[-500(T)80(obias)-600(W)80(e)15(yand,)-688(Andre)-601(Araujo,)-687(Bingyi)-601(Cao,)-688(and)]TJ 21.579 -11.955 Td [(Jack)-533(Sim.)-1266(Google)-533(landmarks)-533(dataset)-533(v2-a)-533(lar)18(ge-)]TJ 0 -11.955 Td [(scale)-219(benchmark)-220(for)-219(instance-le)25(v)15(el)-220(recognition)-219(and)-220(re-)]TJ 0 -11.955 Td [(trie)25(v)25(al.)-551(In)]TJ/F80 9.9626 Tf 44.322 0 Td [(Pr)45(oceedings)-310(of)-309(the)-310(IEEE/CVF)-310(Confer)37(ence)]TJ -44.322 -11.955 Td [(on)-377(Computer)-378(V)74(ision)-377(and)-377(P)80(attern)-378(Reco)10(gnition)]TJ/F76 9.9626 Tf 185.419 0 Td [(,)-409(pages)]TJ -185.419 -11.955 Td [(2575\2262584,)-250(2020.)]TJ
0 g 0 G
0 g 0 G
ET
Q
Q
q
1 0 0 1 0 0 cm
/_Mbf2sx52NzvK6oo7Mi22g Do
Q
endstream
endobj
36 0 obj
<< /Font << /F76 41 0 R /F80 43 0 R >> /ProcSet [ /PDF /Text ] /XObject << /B-h1z9RPOZ22iXiEP1sHoQ 70 0 R /_Mbf2sx52NzvK6oo7Mi22g 71 0 R >> >>
endobj
37 0 obj
<< /Length 5244 >>
stream
q
q
0 g 0 G
0 g 0 G
0 g 0 G
BT
/F76 9.9626 Tf 50.112 710.037 Td [([38])]TJ
0 g 0 G
[-500(Burak)-264(Y)55(ildiz,)-268(Se)15(yran)-264(Khademi,)-268(Ronald)-264(Maria)-265(Siebes,)]TJ 21.579 -11.955 Td [(and)-442(Jan)-443(v)25(an)-442(Gemert.)-976(Amstertime:)-694(A)-442(visual)-443(place)]TJ 0 -11.955 Td [(recognition)-619(benchmark)-618(dataset)-619(for)-618(se)25(v)15(ere)-619(domain)]TJ 0 -11.955 Td [(shift.)]TJ/F80 9.9626 Tf 23.79 0 Td [(arXiv)-250(pr)37(eprint)-250(arXiv:2203.16291)]TJ/F76 9.9626 Tf 132.184 0 Td [(,)-250(2022.)]TJ
0 g 0 G
-177.553 -19.925 Td [([39])]TJ
0 g 0 G
[-500(Martin)-756(Heusel,)-884(Hubert)-756(Ramsauer)40(,)-883(Thomas)-757(Un-)]TJ 21.579 -11.956 Td [(terthiner)40(,)-672(Ber)1(nh)-1(a)1(rd)-588(Nessler)40(,)-671(and)-587(Sepp)-588(Hochreiter)55(.)]TJ 0 -11.955 Td [(Gans)-232(trained)-233(by)-232(a)-232(tw)10(o)-233(time-scale)-232(update)-232(rule)-233(con)40(v)15(er)18(ge)]TJ 0 -11.955 Td [(to)-256(a)-255(local)-256(nash)-256(equilibrium.)]TJ/F80 9.9626 Tf 112.744 0 Td [(Advances)-256(in)-255(neur)15(al)-256(infor)20(-)]TJ -112.744 -11.955 Td [(mation)-250(pr)45(ocessing)-250(systems)]TJ/F76 9.9626 Tf 105.812 0 Td [(,)-250(30,)-250(2017.)]TJ
0 g 0 G
-127.391 -19.925 Td [([40])]TJ
0 g 0 G
[-500(Mik)10<6f07616a>-231(Bi)]TJ 65.577 0.049 Td []TJ -0.832 -0.049 Td [(nk)10(o)25(wski,)-235(Danica)-231(J)-230(Sutherland,)-235(Michael)-231(Ar)20(-)]TJ -43.166 -11.956 Td [(bel,)-460(and)-417(Arthur)-418(Gretton.)-896(Demystifying)-418(mmd)-418(g)5(ans.)]TJ/F80 9.9626 Tf 0 -11.955 Td [(arXiv)-250(pr)37(eprint)-250(arXiv:1801.01401)]TJ/F76 9.9626 Tf 132.183 0 Td [(,)-250(2018.)]TJ
0 g 0 G
-153.762 -19.925 Td [([41])]TJ
0 g 0 G
[-500(Mark)-444(Sabini)-444(and)-444(Gili)-444(Rusak.)-982(P)15(ainti)1(n)-1(g)-444(outside)-444(the)]TJ 21.579 -11.955 Td [(box:)-685(Image)-438(outpainting)-437(with)-438(g)5(ans.)]TJ/F80 9.9626 Tf 155.886 0 Td [(arXiv)-438(pr)37(eprint)]TJ -155.886 -11.955 Td [(arXiv:1808.08483)]TJ/F76 9.9626 Tf 72.777 0 Td [(,)-250(2018.)]TJ
0 g 0 G
-94.356 -19.926 Td [([42])]TJ
0 g 0 G
[-500(Chitw)10(an)-622(Saharia,)-714(W)40(illiam)-622(Chan,)-714(Huiwen)-622(Chang,)]TJ 21.579 -11.955 Td [(Chris)-332(Le)1(e,)-352(Jonathan)-332(Ho,)-352(T)35(im)-331(Salimans,)-352(Da)20(vid)-332(Fleet,)]TJ 0 -11.955 Td [(and)-417(Mohammad)-417(Norouzi.)-894(P)15(alette:)-644(Image-to-image)]TJ 0 -11.955 Td [(dif)25(fusion)-291(models.)-492(In)]TJ/F80 9.9626 Tf 86.003 0 Td [(Special)-291(Inter)37(est)-292(Gr)45(oup)-291(on)-291(Com-)]TJ -86.003 -11.955 Td [(puter)-475(Gr)15(aphics)-474(and)-475(Inter)15(active)-475(T)92(ec)15(hniques)-475(Confer)20(-)]TJ 0 -11.956 Td [(ence)-250(Pr)45(oceedings)]TJ/F76 9.9626 Tf 70.096 0 Td [(,)-250(pages)-250(1\22610,)-250(2022.)]TJ
0 g 0 G
-91.675 -19.925 Td [([43])]TJ
0 g 0 G
[-500(Y)100(anhong)-474(Zeng,)-530(Jianlong)-473(Fu,)-530(Hongyang)-474(Chao,)-530(and)]TJ 21.579 -11.955 Td [(Baining)-269(Guo.)-419(Learning)-269(p)10(yramid-conte)15(xt)-268(encoder)-269(net-)]TJ 0 -11.955 Td [(w)10(ork)-287(for)-287(high-quality)-286(image)-287(inpainting.)-478(In)]TJ/F80 9.9626 Tf 175.291 0 Td [(The)-287(IEEE)]TJ -175.291 -11.955 Td [(Confer)37(ence)-245(on)-246(Computer)-245(V)74(ision)-246(and)-245(P)80(attern)-246(Reco)10(gni-)]TJ 0 -11.956 Td [(tion)]TJ/F76 9.9626 Tf 15.502 0 Td [(,)-250(pages)-250(1486\2261494,)-250(2019.)]TJ
0 g 0 G
-37.081 -19.925 Td [([44])]TJ
0 g 0 G
[-500(Zili)-372(Y)55(i,)-403(Qiang)-372(T)80(ang,)-402(Shek)10(oofeh)-372(Azizi,)-403(Daesik)-372(Jang,)]TJ 21.579 -11.955 Td [(and)-272(Zhan)-271(Xu.)-429(Conte)15(xtual)-272(residual)-272(a)1(g)-1(gr)1(e)14(g)5(ati)1(on)-272(for)-272(ul-)]TJ 0 -11.955 Td [(tra)-250(high-resolution)-250(image)-250(inpainting,)-250(2020.)]TJ
0 g 0 G
-21.579 -19.925 Td [([45])]TJ
0 g 0 G
[-500(Alejandro)-201(L)]TJ 70.341 0.049 Td []TJ -0.832 -0.049 Td [(opez-Cifuentes,)-211(Marcos)-201(Escudero-V)60(i)]TJ 147.369 0.049 Td [<98>]TJ -0.832 -0.049 Td [(nolo,)]TJ -194.467 -11.956 Td [(Jes)]TJ 13.006 0.05 Td []TJ -0.832 -0.05 Td [(us)-354(Besc)]TJ 32.58 0.05 Td []TJ -0.832 -0.05 Td [(os,)-380(and)]TJ 34.978 2.172 Td []TJ -1.938 -2.172 Td [(Alv)25(aro)-354(Garc)]TJ 50.024 0.05 Td []TJ 0.274 -0.05 Td [<11612d4d617274>]TJ 29.604 0.05 Td []TJ 0.274 -0.05 Td [<116e2e>-692(Semantic-)]TJ -157.138 -11.955 Td [(a)15(w)10(are)-743(scene)-743(recognition.)]TJ/F80 9.9626 Tf 127.638 0 Td [(P)80(attern)-743(Reco)10(gnition)]TJ/F76 9.9626 Tf 84.542 0 Td [(,)]TJ -212.18 -11.955 Td [(102:107256,)-250(2020.)]TJ
0 g 0 G
-21.579 -19.925 Td [([46])]TJ
0 g 0 G
[-500(Xinlei)-604(Chen,)-693(Hao)-604(F)15(ang,)-692(Tsung-Y)55(i)-604(Lin,)-693(Ramakr)20(-)]TJ 21.579 -11.955 Td [(ishna)-482(V)111(edant)1(am,)-540(Saurabh)-481(Gupta,)-540(Piotr)-481(Doll)]TJ 185.631 0.049 Td []TJ -0.553 -0.049 Td [(ar)40(,)-539(and)]TJ -185.078 -11.956 Td [(C)-338(La)15(wrence)-339(Zitnick.)-643(Microsoft)-338(coco)-339(captions:)-487(Data)]TJ 0 -11.955 Td [(collection)-611(and)-611(e)25(v)25(aluation)-610(serv)15(er)55(.)]TJ/F80 9.9626 Tf 154.161 0 Td [(arXiv)-611(pr)37(eprint)]TJ -154.161 -11.955 Td [(arXiv:1504.00325)]TJ/F76 9.9626 Tf 72.777 0 Td [(,)-250(2015.)]TJ
0 g 0 G
0 g 0 G
0 g 0 G
0 g 0 G
ET
Q
Q
q
1 0 0 1 0 0 cm
/B-h1z9RPOZ22iXiEP1sHoQ Do
Q
endstream
endobj
38 0 obj
<< /BaseFont /PXOHER+CMR8 /FirstChar 49 /FontDescriptor 72 0 R /LastChar 50 /Subtype /Type1 /ToUnicode 73 0 R /Type /Font /Widths 74 0 R >>
endobj
39 0 obj
<< /BaseFont /ZXXZSM+CMSY9 /FirstChar 102 /FontDescriptor 75 0 R /LastChar 106 /Subtype /Type1 /ToUnicode 76 0 R /Type /Font /Widths 77 0 R >>
endobj
40 0 obj
<< /BaseFont /AHXTXP+NimbusRomNo9L-Medi /Encoding 78 0 R /FirstChar 2 /FontDescriptor 79 0 R /LastChar 180 /Subtype /Type1 /ToUnicode 80 0 R /Type /Font /Widths 81 0 R >>
endobj
41 0 obj
<< /BaseFont /OCDEFA+NimbusRomNo9L-Regu /Encoding 78 0 R /FirstChar 2 /FontDescriptor 82 0 R /LastChar 180 /Subtype /Type1 /ToUnicode 83 0 R /Type /Font /Widths 84 0 R >>
endobj
42 0 obj
<< /BaseFont /UMHVLO+NimbusMonL-Regu /Encoding 78 0 R /FirstChar 44 /FontDescriptor 85 0 R /LastChar 121 /Subtype /Type1 /ToUnicode 86 0 R /Type /Font /Widths 87 0 R >>
endobj
43 0 obj
<< /BaseFont /RDDSTJ+NimbusRomNo9L-ReguItal /Encoding 78 0 R /FirstChar 2 /FontDescriptor 88 0 R /LastChar 122 /Subtype /Type1 /ToUnicode 89 0 R /Type /Font /Widths 90 0 R >>
endobj
44 0 obj
<< /BBox [ 0 0 612 792 ] /Matrix [ 1 0 0 1 0 0 ] /Resources << /Font 91 0 R /ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ] >> /Subtype /Form /Type /XObject /Length 112 /Filter /FlateDecode >>
stream
x-0D|!Ć4V*SGAHuStΧg!뀻/(!c'^=|В~\ٺ3R.rҹ0d|ZuT`o
endstream
endobj
45 0 obj
<< /BBox [ 0 0 612 792 ] /Group << /CS /DeviceRGB /I true /S /Transparency /Type /Group >> /Resources << /ExtGState << /a0 << /CA 1 /ca 1 >> /a1 << /CA 0.5 /ca 0.5 >> >> /XObject << /x6 92 0 R >> >> /Subtype /Form /Type /XObject /Length 382 /Filter /FlateDecode >>
stream
xe[n!Y7P|=Q'y(Ji;#$v0U-V\oZWɽ:5db,6IR6@6Pc"3K+GQ؎3T<9j`UT@
v5ڳtHN;9+7[^S؇:[E"8+sW\qEeКu6Bhsj;)y;Tp0nؚgʦFǃS-="z7ۺ9*&Q~ZPuí}C9嬂8l?Jܪmx%e8w](Rlq鞋t×e:=p=y0`XeJCWxau~uof{`r
endstream
endobj
46 0 obj
<< /BBox [ 0 0 612 792 ] /Matrix [ 1 0 0 1 0 0 ] /Resources << /Font 93 0 R /ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ] >> /Subtype /Form /Type /XObject /Length 112 /Filter /FlateDecode >>
stream
x-0D|K4vҬH01 ;p>=%-nD
=[u-Q)w1A
|N^7F/Tnp
endstream
endobj
47 0 obj
<< /BitsPerComponent 8 /ColorSpace /DeviceRGB /Filter /DCTDecode /Height 602 /Subtype /Image /Type /XObject /Width 1986 /Length 360180 >>
stream
JFIF C
C
Z"
} !1AQa"q2#BR$3br
%&'()*456789:CDEFGHIJSTUVWXYZcdefghijstuvwxyz
w !1AQaq"2B #3Rbr
$4%&'()*56789:CDEFGHIJSTUVWXYZcdefghijstuvwxyz ? S( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( xA>-:=J|fRTe `A~p&<YAycjwrζfܪFQ>lg