%PDF-1.6
%
1 0 obj
<< /Metadata 3 0 R /PageLabels 5 0 R /Pages 8 0 R /Type /Catalog >>
endobj
2 0 obj
<< /Author (Mengmeng Zhu; Guanqun Hou; Xinjia Chen; Jiaxing Xie; Haixian Lu; Jun Che) /Producer (pikepdf 3.1.0) /Subject (IEEE International Conference on Computer Vision Workshops) /Title (Saliency-Guided Transformer Network Combined With Local Embedding for No-Reference Image Quality Assessment) >>
endobj
3 0 obj
<< /Subtype /XML /Type /Metadata /Length 1269 >>
stream
Saliency-Guided Transformer Network Combined With Local Embedding for No-Reference Image Quality AssessmentMengmeng Zhu Guanqun Hou Xinjia Chen Jiaxing Xie Haixian Lu Jun CheIEEE International Conference on Computer Vision Workshops
endstream
endobj
4 0 obj
<< /Type /ObjStm /Length 44 /Filter /FlateDecode /N 2 /First 9 >>
stream
x3U0P0S02ᲱQ+-V()*فŃ]@l v /
endstream
endobj
7 0 obj
<< /Type /ObjStm /Length 80 /Filter /FlateDecode /N 1 /First 4 >>
stream
xP0ಱQw/+Q04PL)VVT0P )Ce! 2PfBYXʂTb;;. CM&
endstream
endobj
9 0 obj
<< /ArtBox [ 0 35.917 595.02 805.943 ] /BleedBox [ 0 35.947 595.02 805.973 ] /Contents [ 19 0 R 20 0 R 21 0 R 22 0 R 23 0 R 24 0 R 25 0 R 26 0 R 27 0 R 28 0 R ] /CropBox [ 0 0 595.22 842 ] /MediaBox [ 0 0 595.22 842 ] /Parent 8 0 R /Resources 30 0 R /Rotate 0 /TrimBox [ 0 35.947 595.02 805.973 ] /Type /Page >>
endobj
10 0 obj
<< /ArtBox [ 0 35.917 595.02 805.943 ] /BleedBox [ 0 35.947 595.02 805.973 ] /Contents [ 45 0 R 46 0 R ] /CropBox [ 0 0 595.22 842 ] /MediaBox [ 0 0 595.22 842 ] /Parent 8 0 R /Resources 48 0 R /Rotate 0 /TrimBox [ 0 35.947 595.02 805.973 ] /Type /Page >>
endobj
11 0 obj
<< /ArtBox [ 0 35.917 595.02 805.943 ] /BleedBox [ 0 35.947 595.02 805.973 ] /Contents [ 49 0 R 50 0 R ] /CropBox [ 0 0 595.22 842 ] /MediaBox [ 0 0 595.22 842 ] /Parent 8 0 R /Resources 52 0 R /Rotate 0 /TrimBox [ 0 35.947 595.02 805.973 ] /Type /Page >>
endobj
12 0 obj
<< /ArtBox [ 0 35.917 595.02 805.943 ] /BleedBox [ 0 35.947 595.02 805.973 ] /Contents [ 58 0 R 59 0 R ] /CropBox [ 0 0 595.22 842 ] /MediaBox [ 0 0 595.22 842 ] /Parent 8 0 R /Resources 61 0 R /Rotate 0 /TrimBox [ 0 35.947 595.02 805.973 ] /Type /Page >>
endobj
13 0 obj
<< /ArtBox [ 0 35.917 595.02 805.943 ] /BleedBox [ 0 35.947 595.02 805.973 ] /Contents [ 68 0 R 69 0 R ] /CropBox [ 0 0 595.22 842 ] /MediaBox [ 0 0 595.22 842 ] /Parent 8 0 R /Resources 71 0 R /Rotate 0 /TrimBox [ 0 35.947 595.02 805.973 ] /Type /Page >>
endobj
14 0 obj
<< /ArtBox [ 0 35.917 595.02 805.943 ] /BleedBox [ 0 35.947 595.02 805.973 ] /Contents [ 79 0 R 80 0 R ] /CropBox [ 0 0 595.22 842 ] /MediaBox [ 0 0 595.22 842 ] /Parent 8 0 R /Resources 82 0 R /Rotate 0 /TrimBox [ 0 35.947 595.02 805.973 ] /Type /Page >>
endobj
15 0 obj
<< /ArtBox [ 0 35.917 595.02 805.943 ] /BleedBox [ 0 35.947 595.02 805.973 ] /Contents [ 84 0 R 85 0 R ] /CropBox [ 0 0 595.22 842 ] /MediaBox [ 0 0 595.22 842 ] /Parent 8 0 R /Resources 87 0 R /Rotate 0 /TrimBox [ 0 35.947 595.02 805.973 ] /Type /Page >>
endobj
16 0 obj
<< /ArtBox [ 0 35.917 595.02 805.943 ] /BleedBox [ 0 35.947 595.02 805.973 ] /Contents [ 92 0 R 93 0 R ] /CropBox [ 0 0 595.22 842 ] /MediaBox [ 0 0 595.22 842 ] /Parent 8 0 R /Resources 95 0 R /Rotate 0 /TrimBox [ 0 35.947 595.02 805.973 ] /Type /Page >>
endobj
17 0 obj
<< /ArtBox [ 0 35.917 595.02 805.943 ] /BleedBox [ 0 35.947 595.02 805.973 ] /Contents [ 96 0 R 97 0 R ] /CropBox [ 0 0 595.22 842 ] /MediaBox [ 0 0 595.22 842 ] /Parent 8 0 R /Resources 99 0 R /Rotate 0 /TrimBox [ 0 35.947 595.02 805.973 ] /Type /Page >>
endobj
18 0 obj
<< /ArtBox [ 0 35.917 595.02 805.943 ] /BleedBox [ 0 35.947 595.02 805.973 ] /Contents [ 102 0 R 103 0 R ] /CropBox [ 0 0 595.22 842 ] /MediaBox [ 0 0 595.22 842 ] /Parent 8 0 R /Resources 105 0 R /Rotate 0 /TrimBox [ 0 35.947 595.02 805.973 ] /Type /Page >>
endobj
19 0 obj
<< /Length 1299 >>
stream
/GS1 gs
BT
/F1 1 Tf
13.9482 0 0 13.9482 61.8909 692.394 Tm
/Cs8 cs 1 scn
0 Tc
0 Tw
[(Saliency-Guided)-250(T)74(ransf)25(ormer)-250(Netw)10(ork)-250(combined)-250(with)-250(Local)-250(Embedding)-250(f)25(o)0(r)]TJ
7.6383 -1.25 TD
[(No-Refer)18(ence)-250(Image)-250(Quality)-250(Assessment)]TJ
/F2 1 Tf
11.6235 0 0 11.6235 78.8607 628.4659 Tm
[(Mengmeng)-250(Zhu)]TJ
7.749 0 0 7.749 153.4363 632.6829 Tm
(1)Tj
11.6235 0 0 11.6235 157.7953 628.4659 Tm
[(,)-250(Guanqun)-250(Hou)]TJ
7.749 0 0 7.749 229.1404 632.6829 Tm
(2)Tj
11.6235 0 0 11.6235 233.4994 628.4659 Tm
[(,)-250(Xinjia)-250(Chen)]TJ
7.749 0 0 7.749 295.8116 632.6829 Tm
(2)Tj
11.6235 0 0 11.6235 300.1706 628.4659 Tm
[(,)-250(Jiaxing)-250(Xie)]TJ
7.749 0 0 7.749 359.2527 632.6829 Tm
(2)Tj
11.6235 0 0 11.6235 363.6117 628.4659 Tm
[(,)-250(Haixian)-250(Lu)]TJ
7.749 0 0 7.749 422.0418 632.6829 Tm
(2)Tj
11.6235 0 0 11.6235 426.4008 628.4659 Tm
[(,)-250(and)-250(Jun)-250(Che)]TJ
7.749 0 0 7.749 489.679 632.6829 Tm
(2,*)Tj
-43.6512 -2.9999 TD
(1)Tj
11.6235 0 0 11.6235 155.7843 605.2178 Tm
[(Uni)25(v)15(ersity)-250(of)-250(Electronic)-250(Science)-250(and)-250(T)70(echnology)-250(of)-250(China)]TJ
7.749 0 0 7.749 220.9934 595.8759 Tm
(2)Tj
11.6235 0 0 11.6235 225.3514 591.6578 Tm
[(Hikvision)-250(Research)-250(Institute)]TJ
/F1 1 Tf
endstream
endobj
20 0 obj
<< /Length 1379 >>
stream
-7.1757 -3.9307 TD
(Abstract)Tj
/F3 1 Tf
9.6862 0 0 9.6862 60.345 521.5357 Tm
[(No-Refer)37(ence)-769(Ima)10(g)10(e)-769(Quality)-769(Assessment)-769(\(NR-IQA\))]TJ
-1.2 -1.2 TD
[(methods)-281(based)-281(on)-281(V)74(ision)-281(T)55(r)15(ansformer)-281(have)-281(r)37(ecently)-281(dr)15(awn)]TJ
T*
[(muc)15(h)-638(attention)-637(for)-638(their)-638(superior)-638(performance)14(.)-1473(Un-)]TJ
0 -1.2002 TD
[(fortunately)55(,)-565(being)-502(a)-502(crude)-501(combination)-502(of)-502(NR-IQA)-502(and)]TJ
0 -1.2 TD
[(T)55(r)15(ansformer)111(,)-438(the)30(y)-401(can)-401(har)37(dly)-401(tak)10(e)-400(the)-401(advanta)10(g)10(e)-401(of)-401(their)]TJ
T*
[(str)37(engths.)-1000(In)-480(this)-481(paper)111(,)-537(w)0(e)-480(p)0(r)45(opose)-480(a)-481(n)0(o)10(vel)-480(Saliency-)]TJ
T*
[(Guided)-233(T)55(r)15(ansformer)-232(Network)-233(combined)-232(with)-233(Local)-233(Embed-)]TJ
T*
[(ding)-446(\(T)55(r)15(anSLA\))-446(for)-446(No-Refer)37(ence)-446(Ima)10(g)10(e)-447(Quality)-446(Assess-)]TJ
T*
[(ment.)-674(Our)-371(T)55(r)15(anSLA)-371(inte)40(gr)15(ates)-372(dif)18(fer)37(ent-le)15(vel)-372(information)]TJ
0 -1.2002 TD
[(for)-262(a)-262(r)46(ob)20(ust)-261(r)37(epr)37(esentation.)-345(Existed)-262(r)37(esear)37(c)15(hes)-262(have)-262(shown)]TJ
0 -1.2 TD
[(that)-478(the)-478(human)-478(vision)-478(system)-477.9(concentr)15(ates)-478(mor)37(e)-478(o)0(n)-478(the)]
endstream
endobj
21 0 obj
<< /Length 1404 >>
stream
TJ
T*
[(Re)40(gion-of-inter)37(est)-316(\(RoI\))-316(when)-316(assessing)-316(the)-316(ima)10(g)10(e)-316(quality)55(.)]TJ
T*
[(Thus)-347(we)-348(combine)-347(saliency)-348(pr)37(ediction)-347(with)-347(T)55(r)15(ansformer)-348(to)]TJ
T*
[(guide)-428(the)-428(model)-428(highlight)-426.9(the)-428.1(RoI)-427.9(when)-428(a)10(g)10(g)0(r)37(e)40(gating)-428(the)]TJ
0 -1.2001 TD
[(global)-390(information.)-729(Besides,)-424.9(we)-390(import)-390(local)-390(embedding)]TJ
0 -1.2 TD
[(for)-243(T)55(r)15(ansformer)-243(with)-243(gr)15(adient)-242(map.)-308(Since)-243(the)-243(gr)15(adient)-243(map)]TJ
0 -1.2001 TD
[(focuses)-291(on)-291(e)20(xtr)15(acting)-291(structur)37(ed)-291(featur)37(e)-291(i)0(n)-291(detail,)-302(it)-291(can)-291(be)]TJ
0 -1.2 TD
[(used)-311(as)-311(a)-312(supplement)-311(to)-311(of)18(fer)-311(local)-311(information)-311(for)-312(T)55(r)15(ans-)]TJ
T*
[(former)111(.)-567(Then,)-357(the)-335(local)-336(and)-335(non-local)-336(information)-335(can)-336(be)]TJ
T*
[(utilized.)-536(Mor)37(eo)10(ver)111(,)-344(t)0(o)-325(acceler)15(ate)-325(the)-325(a)10(g)10(g)0(r)37(e)40(gation)-325(of)-326(infor)20(-)]TJ
0 -1.2002 TD
[(mation)-232(fr)45(om)-231(all)-232(tok)10(ens,)-235(we)-232(intr)45(oduce)-231(a)-232(Boosting)-232(Inter)15(action)]TJ
0 -1.2 TD
[(Module)-280(\(BIM\))-281(to)-280(enhance)-281(featur)37(e)-280(a)10(g)10(g)0(r)37(e)40(gation.)-401(BIM)-281(for)37(ces)]
endstream
endobj
22 0 obj
<< /Length 1312 >>
stream
TJ
T*
[(patc)15(h)-317(tok)10(ens)-317(to)-317(inter)15(act)-316(better)-317(with)-317(class)-317(tok)10(ens)-317(at)-317(all)-317(le)15(v-)]TJ
0 -1.2001 TD
[(els.)-611(Experiments)-349.9(on)-351(two)-350(lar)37(g)10(e-scale)-350(NR-IQA)-351(benc)15(hmarks)]TJ
T*
[(demonstr)15(ate)-316(that)-316(our)-316(method)-316(signi\223cantly)-316(outperforms)-316(the)]TJ
T*
(state-of-the-art.)Tj
/F1 1 Tf
11.6235 0 0 11.6235 48.7217 192.517 Tm
[(1.)-250(Intr)18(oduction)]TJ
/F2 1 Tf
9.6862 0 0 9.6862 60.345 172.927 Tm
[(W)40(ith)-267(the)-267(rapid)-267(gro)25(wth)-267(of)-267(digital)-267(image)-267(and)-267(transmission)]TJ
-1.2 -1.2001 TD
[(technology)65(,)-357(No-Reference)-336(Image)-336(quality)-336(assessment)-336(\(NR-)]TJ
T*
[(IQA\))-339(has)-339(a)-340(wide)-339(range)-339(of)-339(applications)-339(in)-340(image)-339(compres-)]TJ
0 -1.2 TD
[(sion,)-385(image)-357(reco)15(v)15(ery)65(,)-385(and)-358(video)-357(surv)15(eillance,)-385(etc.)-633([38,)-358(2,)]TJ
0 -1.2001 TD
[(8,)-288(22].)-426(The)-288(subjecti)25(v)15(e)-289(quality)-288(assessment)-289(measure)-288(the)-289(per)20(-)]TJ
0 -1.2 TD
[(cei)25(v)15(e)0(d)-414(quality)-414(by)-414(mean)-414(opinion)-415(scores)-414(\(MOS\))-414(from)-414(col-)]TJ
ET
0 scn
300.294 449.943 235.811 105.712 re
f
q
1 i
302.669 459.409 55.076 41.307 re
endstream
endobj
23 0 obj
<< /Length 1308 >>
stream
W n
q
55.166 0 0 41.412 302.634 459.354 cm
/Im1 Do
Q
Q
q
1 i
303.202 514.038 55.076 41.307 re
W n
q
55.166 0 0 41.352 303.174 514.014 cm
/Im2 Do
Q
Q
BT
/TT2 1 Tf
5.7314 0 0 5.7314 312.3586 506.9837 Tm
1 scn
-.0001 Tc
[(\(a\))-224.2(M)-.1(OS=1.19)]TJ
ET
q
1 i
419.223 514.038 55.076 41.307 re
W n
q
55.166 0 0 41.352 419.214 514.014 cm
/Im3 Do
Q
Q
q
1 i
478.206 514.038 55.076 41.307 re
W n
q
55.166 0 0 41.352 478.194 514.014 cm
/Im4 Do
Q
Q
q
1 i
361.652 514.038 55.076 41.307 re
W n
q
55.166 0 0 41.352 361.614 514.014 cm
/Im5 Do
Q
Q
q
1 i
478.206 459.754 55.076 41.306 re
W n
q
55.166 0 0 41.412 478.194 459.714 cm
/Im6 Do
Q
Q
q
1 i
361.652 459.754 55.076 41.306 re
W n
q
55.166 0 0 41.412 361.614 459.714 cm
/Im7 Do
Q
Q
q
1 i
419.223 459.754 55.076 41.306 re
W n
q
55.166 0 0 41.412 419.214 459.714 cm
/Im8 Do
Q
Q
BT
5.7314 0 0 5.7314 313.8966 453.6705 Tm
0 Tc
[(\(e\) MOS=3.82)-4584.5(\(f)-15.8(\) MOS=3.37)-14974.9(\(f)-15.8(\) MOS=2.90)]TJ
20.2447 0 TD
.0004 Tc
.0008 Tw
(\(g\) MOS=3.00)Tj
ET
0 .99335 1 0 K
0 J 0 j .215 w 10 M [] 0 d
/GS2 gs
1 i
327.341 480.398 18.227 11.274 re
S
.62066 .10959 .98141 .004211 K
.323 w
306.292 464.925 18.226 11.274 re
S
0 .99335 1 0 K
.215 w
375.946 473.307 11.91 11.274 re
S
.62066 .10959 .98141 .004211 K
395.498 474.357 11.152 10.43 re
S
0 .99335 1 0 K
endstream
endobj
24 0 obj
<< /Length 1382 >>
stream
444.05 484.736 11.911 11.256 re
S
.62066 .10959 .98141 .004211 K
429.559 484.082 11.91 11.256 re
S
0 .99335 1 0 K
485.444 470.054 21.737 11.257 re
S
.62066 .10959 .98141 .004211 K
508.782 484.082 11.91 11.256 re
S
BT
5.7314 0 0 5.7314 429.0408 506.9837 Tm
-.0001 Tc
0 Tw
[(\(c\))-225.5(M)-1.6(OS=1.49)]TJ
-10.1789 0 TD
.0002 Tc
[(\(b\))-225.3(M)-.7(OS=1.15)-14753.6(\().1(d\))-225.3(MOS=2.02)]TJ
/F2 1 Tf
8.7176 0 0 8.7176 300.2936 439.2885 Tm
0 Tc
[(Figure)-236.9(1.)-542(Examples)-237(of)-236(wild)-237(images)-236(and)-237(their)-237(corresponding)-236(mean)]TJ
0 -1.2222 TD
[(opinion)-233(scores)-232.9(\(MOS\))-233.1(with)-233(dif)25(ferent)-232(image)-233(quality)-233(le)25(v)15(els.)-305(The)-232(red)]TJ
T*
[(box)-249.9(is)-250(the)-250(clear)-250(RoI,)-250(and)-250(the)-250(green)-250(box)-250(is)-250(the)-250(blur)-250(background.)]TJ
9.6862 0 0 9.6862 300.2936 383.7764 Tm
[(lected)-197.9(subjecti)25(v)15(e)-198(ratings.)-293(Ho)25(we)25(v)15(e)0(r)40(,)-208(subjecti)25(v)15(e)-199(NR-IQA)-198(can-)]TJ
0 -1.2 TD
[(not)-254(be)-253(used)-254(in)-254(v)25(arious)-253(scenarios)-254(due)-253(to)-254(labor)-254(consume.)-321(The)]TJ
T*
[(de)25(v)15(elopment)-367(of)-367(automatic,)-397(rob)20(ust,)-396(and)-368(objecti)25(v)15(e)-368(NR-IQA)]TJ
0 -1.2001 TD
[(technology)-398.9(is)-399(of)-400(practical)-400(signi\223cance)-399([39,)-399(35].)-758(In)-399(early)]
endstream
endobj
25 0 obj
<< /Length 1346 >>
stream
TJ
T*
[(approaches,)-271(majority)-266.9(ef)25(forts)-267(focus)-267(on)-267(CNN-based)-267(NR-IQA)]TJ
0 -1.2 TD
[(methods)-292.9([45,)-292(34,)-292.9(40])-293(due)-292(to)-293(the)-292(po)25(werful)-293(feature)-293(represen-)]TJ
T*
[(tation)-306.9(ability)-307(of)-306(con)40(v)20(olution)-307(neural)-307(netw)10(ork)-307(\(CNN\).)-307(These)]TJ
T*
[(methods)-381(are)-381(mainly)-380.9(based)-382(on)-381(three)-381(ideas.)-703(1\))-381(The)-382<93727374>-381(is)]TJ
T*
[(to)-425(directly)-424(predict)-425(the)-425(MOS)-425(by)-424(establishing)-425(a)-425(r)0(e)15(gression)]TJ
T*
[(model.)-638.9(2\)The)-360(second)-360(is)-360(simulating)-360(Human)-360(visual)-360(system)]TJ
0 -1.2002 TD
[(\(HVS\))-340.9(to)-341(obtain)-341(the)-341(corresponding)-341(\(reference)-342(image\))-341(pre-)]TJ
0 -1.2 TD
[(diction)-301.9([29,)-301.9(23].)-466(3\)The)-302(third)-301.9(is)-302(using)-302(a)-302(multi-task)-302(learning)]TJ
T*
[(strate)15(gy)-250([24,)-250(43].)]TJ
1.2 -1.368 TD
[(Although)-528(these)-527.9(methods)-528(ha)20(v)15(e)-528(achie)25(v)15(e)0(d)-529(good)-528(perfor)20(-)]TJ
-1.2 -1.2001 TD
[(mance)-375(based)-374(on)-375(arti\223cial)-375(distortions,)-405(a)-375(signi\223cant)-375(perfor)20(-)]TJ
T*
[(mance)-296.9(drop)-298(can)-297.1(still)-297.9(be)-297(observ)15(ed)-297(in)-298(the)-297(authentic)-298(distorted)]
endstream
endobj
26 0 obj
<< /Length 982 >>
stream
TJ
T*
[(application)-226.9(due)-226(to)-227(distortion)-226(di)25(v)15(ersity)-227(and)-226(content)-227(v)25(ariation.)]TJ
T*
[(T)80(o)-343(address)-343(the)-343(problem,)-366(methods)-343([34,)-343(40])-343(combined)-343(v)25(ari-)]TJ
T*
[(ous)-277.9(features)-278(at)-278(multi-scale)-278(h)0(a)20(v)15(e)-278(been)-278(proposed.)-394(HyperIQA)]TJ
T*
[([34])-245(proposed)-245(a)-244(local)-245(distortion)-245(a)15(w)10(are)-244(module)-245(to)-245(e)15(xtract)-245(lo-)]TJ
T*
[(cal)-296(features)-296(from)-295.9(multi-scale)-296(to)-296(handle)-296(distortion)-296(di)25(v)15(ersity)65(.)]TJ
T*
[(CaHDC)-283([40])-282(utilized)-283(a)-283(cascaded)-283(architecture)-282(to)-283(deri)25(v)15(e)-283(fea-)]TJ
T*
[(tures)-280(at)-279(dif)25(ferent)-280(scales.)-399(Unfortunately)65(,)-287(the)15(y)-280(g)5(a)20(v)15(e)-280(the)-280(same)]TJ
T*
[(de)15(gree)-240(of)-240(attention)-240(to)-240(each)-240(part)-239(of)-240(the)-240(image.)-307(This)-240(is)-240(unreli-)]TJ
ET
endstream
endobj
27 0 obj
<< /Length 46 >>
stream
q
1 0 0 1 0 0 cm
/6s420kz2F4bgT_cNanLQTQ Do
Q
endstream
endobj
28 0 obj
<< /Length 46 >>
stream
q
1 0 0 1 0 0 cm
/RV3APOJblMp3MmezYvpLgA Do
Q
endstream
endobj
29 0 obj
<< /Type /ObjStm /Length 1812 /Filter /FlateDecode /N 15 /First 112 >>
stream
xkOJWsTqT! - Ph{T9&qq`;@睵q$wٙk)l'%-CW!n0M/Luď
08`[!4lzFa`P<`q){,ETGFqc7]OdT.o*l*^]Ylx,Ŀ(Y>NT>
Q9;qduQ?Z^\]Ө