%PDF-1.6
%
1 0 obj
<< /Metadata 3 0 R /Names 4 0 R /OpenAction 5 0 R /PageMode /UseNone /Pages 6 0 R /Type /Catalog >>
endobj
2 0 obj
<< /Author (Yuecong Min; Aiming Hao; Xiujuan Chai; Xilin Chen) /Producer (pikepdf 3.1.0) /Subject (IEEE International Conference on Computer Vision) /Title (Visual Alignment Constraint for Continuous Sign Language Recognition) >>
endobj
3 0 obj
<< /Subtype /XML /Type /Metadata /Length 1165 >>
stream
Visual Alignment Constraint for Continuous Sign Language RecognitionYuecong Min Aiming Hao Xiujuan Chai Xilin ChenIEEE International Conference on Computer Vision
endstream
endobj
4 0 obj
<< /Dests 7 0 R >>
endobj
5 0 obj
<< /D [ 8 0 R /Fit ] /S /GoTo >>
endobj
6 0 obj
<< /Count 10 /Kids [ 9 0 R 10 0 R ] /Type /Pages >>
endobj
7 0 obj
<< /Kids [ 11 0 R 12 0 R 13 0 R ] /Limits [ (Doc-Start) (table.6) ] >>
endobj
8 0 obj
<< /Annots [ 14 0 R 15 0 R 16 0 R 17 0 R 18 0 R 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 29 0 R 30 0 R 31 0 R 32 0 R ] /Contents [ 33 0 R 34 0 R 35 0 R ] /Group 36 0 R /MediaBox [ 0 0 612 792 ] /Parent 9 0 R /Resources 37 0 R /Type /Page >>
endobj
9 0 obj
<< /Count 6 /Kids [ 8 0 R 38 0 R 39 0 R 40 0 R 41 0 R 42 0 R ] /Parent 6 0 R /Type /Pages >>
endobj
10 0 obj
<< /Count 4 /Kids [ 43 0 R 44 0 R 45 0 R 46 0 R ] /Parent 6 0 R /Type /Pages >>
endobj
11 0 obj
<< /Kids [ 47 0 R 48 0 R 49 0 R 50 0 R 51 0 R 52 0 R ] /Limits [ (Doc-Start) (cite.li2020word) ] >>
endobj
12 0 obj
<< /Kids [ 53 0 R 54 0 R 55 0 R 56 0 R 57 0 R 58 0 R ] /Limits [ (cite.msasl) (page.8) ] >>
endobj
13 0 obj
<< /Kids [ 59 0 R 60 0 R 61 0 R 62 0 R 63 0 R 64 0 R ] /Limits [ (page.9) (table.6) ] >>
endobj
14 0 obj
<< /A << /D (cite.dreuw2007speech) /S /GoTo >> /Border [ 0 0 0 ] /C [ 0 1 0 ] /H /I /Rect [ 256.719 179.936 268.675 188.862 ] /Subtype /Link /Type /Annot >>
endobj
15 0 obj
<< /A << /D (cite.ong2005automatic) /S /GoTo >> /Border [ 0 0 0 ] /C [ 0 1 0 ] /H /I /Rect [ 272.086 179.936 284.041 188.862 ] /Subtype /Link /Type /Annot >>
endobj
16 0 obj
<< /A << /D (cite.sandler2006sign) /S /GoTo >> /Border [ 0 0 0 ] /C [ 0 1 0 ] /H /I /Rect [ 191.647 168.12 203.602 176.907 ] /Subtype /Link /Type /Annot >>
endobj
17 0 obj
<< /A << /D (cite.bragg2019sign) /S /GoTo >> /Border [ 0 0 0 ] /C [ 0 1 0 ] /H /I /Rect [ 452.946 367.07 459.92 375.857 ] /Subtype /Link /Type /Annot >>
endobj
18 0 obj
<< /A << /D (cite.koller2019weakly) /S /GoTo >> /Border [ 0 0 0 ] /C [ 0 1 0 ] /H /I /Rect [ 400.748 283.309 412.703 292.171 ] /Subtype /Link /Type /Annot >>
endobj
19 0 obj
<< /A << /D (cite.koller2016deephand) /S /GoTo >> /Border [ 0 0 0 ] /C [ 0 1 0 ] /H /I /Rect [ 416.526 283.165 428.481 292.171 ] /Subtype /Link /Type /Annot >>
endobj
20 0 obj
<< /A << /D (cite.koller2017re) /S /GoTo >> /Border [ 0 0 0 ] /C [ 0 1 0 ] /H /I /Rect [ 432.305 283.245 444.26 292.171 ] /Subtype /Link /Type /Annot >>
endobj
21 0 obj
<< /A << /D (cite.graves2006connectionist) /S /GoTo >> /Border [ 0 0 0 ] /C [ 0 1 0 ] /H /I /Rect [ 324.412 259.334 336.367 268.26 ] /Subtype /Link /Type /Annot >>
endobj
22 0 obj
<< /A << /D (cite.camgoz2017subunets) /S /GoTo >> /Border [ 0 0 0 ] /C [ 0 1 0 ] /H /I /Rect [ 337.923 211.513 344.897 220.44 ] /Subtype /Link /Type /Annot >>
endobj
23 0 obj
<< /A << /D (cite.cheng2020fully) /S /GoTo >> /Border [ 0 0 0 ] /C [ 0 1 0 ] /H /I /Rect [ 347.885 211.513 354.859 220.44 ] /Subtype /Link /Type /Annot >>
endobj
24 0 obj
<< /A << /D (cite.cui2017recurrent) /S /GoTo >> /Border [ 0 0 0 ] /C [ 0 1 0 ] /H /I /Rect [ 357.848 211.513 364.822 220.44 ] /Subtype /Link /Type /Annot >>
endobj
25 0 obj
<< /A << /D (cite.cui2019deep) /S /GoTo >> /Border [ 0 0 0 ] /C [ 0 1 0 ] /H /I /Rect [ 367.811 211.434 374.784 220.44 ] /Subtype /Link /Type /Annot >>
endobj
26 0 obj
<< /A << /D (cite.koller2019weakly) /S /GoTo >> /Border [ 0 0 0 ] /C [ 0 1 0 ] /H /I /Rect [ 377.773 211.578 389.728 220.44 ] /Subtype /Link /Type /Annot >>
endobj
27 0 obj
<< /A << /D (cite.niustochastic) /S /GoTo >> /Border [ 0 0 0 ] /C [ 0 1 0 ] /H /I /Rect [ 392.717 211.513 404.672 220.44 ] /Subtype /Link /Type /Annot >>
endobj
28 0 obj
<< /A << /D (cite.wang2018connectionist) /S /GoTo >> /Border [ 0 0 0 ] /C [ 0 1 0 ] /H /I /Rect [ 407.661 211.513 419.616 220.44 ] /Subtype /Link /Type /Annot >>
endobj
29 0 obj
<< /A << /D (cite.cui2019deep) /S /GoTo >> /Border [ 0 0 0 ] /C [ 0 1 0 ] /H /I /Rect [ 456.789 187.452 463.763 196.458 ] /Subtype /Link /Type /Annot >>
endobj
30 0 obj
<< /A << /D (cite.pu2019iterative) /S /GoTo >> /Border [ 0 0 0 ] /C [ 0 1 0 ] /H /I /Rect [ 467.547 187.452 479.502 196.458 ] /Subtype /Link /Type /Annot >>
endobj
31 0 obj
<< /A << /D (cite.cheng2020fully) /S /GoTo >> /Border [ 0 0 0 ] /C [ 0 1 0 ] /H /I /Rect [ 505.878 115.801 512.852 124.727 ] /Subtype /Link /Type /Annot >>
endobj
32 0 obj
<< /A << /D (cite.niustochastic) /S /GoTo >> /Border [ 0 0 0 ] /C [ 0 1 0 ] /H /I /Rect [ 516.558 115.801 528.514 124.727 ] /Subtype /Link /Type /Annot >>
endobj
33 0 obj
<< /Length 10934 >>
stream
0 g 0 G
0 g 0 G
0 g 0 G
0 g 0 G
0 g 0 G
BT
/F65 14.3462 Tf 75.471 675.067 Td [(V)37(isual)-250(Alignment)-250(Constraint)-250(f)25(or)-250(Continuous)-250(Sign)-250(Language)-250(Recognition)]TJ/F66 11.9552 Tf 80.801 -37.858 Td [(Y)111(uecong)-250(Min)]TJ/F66 7.9701 Tf 64.749 4.339 Td [(1,2)]TJ/F66 11.9552 Tf 10.461 -4.339 Td [(,)-250(Aiming)-250(Hao)]TJ/F66 7.9701 Tf 65.418 4.339 Td [(1,2)]TJ/F66 11.9552 Tf 10.461 -4.339 Td [(,)-250(Xiujuan)-250(Chai)]TJ/F66 7.9701 Tf 70.068 4.339 Td [(3)]TJ/F66 11.9552 Tf 4.484 -4.339 Td [(,)-250(Xilin)-250(Chen)]TJ/F66 7.9701 Tf 58.783 4.339 Td [(1,2)]TJ -348.825 -13.959 Td [(1)]TJ/F66 11.9552 Tf 4.483 -4.338 Td [(K)25(e)15(y)-250(Lab)-250(of)-250(Intelligent)-250(Information)-250(Processing)-250(of)-250(Chinese)-250(Academy)-250(of)-250(Sciences)-250(\(CAS\),)]TJ 50.253 -13.948 Td [(Institute)-250(of)-250(Computing)-250(T)70(echnology)65(,)-250(CAS,)-250(Beijing,)-250(100190,)-250(China)]TJ/F66 7.9701 Tf -9.427 -9.62 Td [(2)]TJ/F66 11.9552 Tf 4.484 -4.338 Td [(Uni)25(v)15(ersity)-250(of)-250(Chinese)-250(Academy)-250(of)-250(Sciences,)-250(Beijing,)-250(100049,)-250(China)]TJ/F66 7.9701 Tf -85.574 -9.62 Td [(3)]TJ/F66 11.9552 Tf 4.483 -4.338 Td [(Agricultural)-250(Information)-250(Institute,)-250(Chinese)-250(Academy)-250(of)-250(Agricultural)-250(Sciences,)-250(Beijing,)-250(100081,)-250(China)]TJ/F61 8.9664 Tf 34.101 -13.948 Td [(f)]TJ/F68 8.9664 Tf 4.607 0 Td [(yuecong.min,aiming.hao)]TJ/F61 8.9664 Tf 118.356 0 Td [(g)]TJ/F68 8.9664 Tf 4.608 0 Td [(@vipl.ict.ac.cn,)-600(chaixiujuan@caas.cn,)-600(xlchen@ict.ac.cn)]TJ
0 g 0 G
0 g 0 G
/F65 11.9552 Tf -76.25 -41.046 Td [(Abstract)]TJ/F70 9.9626 Tf -83.928 -24.631 Td [(V)74(ision-based)-543(Continuous)-543(Sign)-544(Langua)10(g)10(e)-543(Reco)10(gnition)]TJ -11.955 -11.955 Td [(\(CSLR\))-386(aims)-386(to)-385(r)37(eco)10(gnize)-386(unse)40(gmented)-386(signs)-386(fr)45(om)-386(ima)10(g)10(e)]TJ 0 -11.955 Td [(str)37(eams.)-739(Over\002tting)-393(is)-393(one)-393(of)-394(the)-393(most)-393(critical)-393(pr)45(oblems)]TJ 0 -11.956 Td [(in)-313(CSLR)-313(tr)15(aining)10(,)-328(and)-313(pr)37(e)15(vious)-313(works)-313(show)-313(that)-313(the)-313(iter)15(a-)]TJ 0 -11.955 Td [(tive)-286(tr)15(aining)-286(sc)15(heme)-287(can)-286(partially)-286(solve)-286(this)-286(pr)45(oblem)-287(while)]TJ 0 -11.955 Td [(also)-254(costing)-254(mor)37(e)-254(tr)15(aining)-254(time)15(.)-322(In)-254(this)-255(study)55(,)-255(we)-254(r)37(e)15(visit)-254(the)]TJ 0 -11.955 Td [(iter)15(ative)-361(tr)15(aining)-361(sc)15(heme)-362(in)-361(r)37(ecent)-361(CSLR)-361(works)-361(and)-362(r)37(eal-)]TJ 0 -11.955 Td [(ize)-251(that)-252(suf)18(\002cient)-251(tr)15(aining)-252(of)-251(the)-252(feat)1(u)-1(r)38(e)-252(e)20(xtr)15(actor)-251(is)-252(critical)]TJ 0 -11.956 Td [(to)-356(solving)-355(the)-356(o)10(ver\002tting)-356(pr)45(oblem.)-627(Ther)37(efor)37(e)10(,)-382(we)-356(pr)45(opose)]TJ 0 -11.955 Td [(a)-376(V)74(isual)-375(Alignment)-376(Constr)15(aint)-375(\(V)60(A)30(C\))-376(to)-375(enhance)-376(the)-376(fea-)]TJ 0 -11.955 Td [(tur)37(e)-291(e)20(xtr)15(actor)-291(with)-291(alignment)-291(supervision.)-433(Speci\002cally)55(,)-301(the)]TJ 0 -11.955 Td [(pr)45(oposed)-276(V)60(A)30(C)-276(comprises)-276(two)-275(auxiliary)-276(losses:)-362(one)-276(focuses)]TJ 0 -11.955 Td [(on)-367(visual)-367(featur)37(es)-367(only)55(,)-396(and)-367(the)-367(other)-367(enfor)37(ces)-367(pr)37(ediction)]TJ 0 -11.955 Td [(alignment)-295(between)-295(the)-295(featur)37(e)-295(e)20(xtr)15(actor)-295(and)-296(the)-295(alignment)]TJ 0 -11.956 Td [(module)15(.)-293(Mor)37(eo)10(ver)111(,)-211(we)-200(pr)45(opose)-200(two)-200(metrics)-200(to)-201(r)37(e\003ect)-200(o)10(ver\002t-)]TJ 0 -11.955 Td [(ting)-258(by)-259(measuring)-258(the)-259(pr)37(ediction)-258(inconsistency)-258(between)-259(the)]TJ 0 -11.955 Td [(featur)37(e)-305(e)20(xtr)15(actor)-304(and)-305(the)-305(alignment)-305(module)15(.)-474(Experimental)]TJ 0 -11.955 Td [(r)37(esults)-201(on)-201(two)-201(c)15(hallenging)-201(CSLR)-201(datasets)-201(show)-201(that)-201(the)-201(pr)45(o-)]TJ 0 -11.955 Td [(posed)-238(V)60(A)30(C)-237(mak)10(es)-238(CSLR)-238(networks)-237(end-to-end)-238(tr)15(ainable)-238(and)]TJ 0 -11.955 Td [(ac)15(hie)15(ves)-250(competitive)-250(performance)15(.)]TJ/F65 11.9552 Tf 0 -38.028 Td [(1.)-250(Intr)18(oduction)]TJ/F66 9.9626 Tf 11.955 -19.649 Td [(Sign)-590(Language)-590(is)-589(a)-590(complete)-590(and)-590(natural)-590(language)]TJ -11.955 -11.955 Td [(that)-270(con)40(v)15(e)15(ys)-270(information)-269(through)-270(both)-270(manual)-270(components)]TJ 0 -11.956 Td [(\(hand/arm)-479(gestures\))-479(and)-479(non-manual)-480(components)-479(\(f)10(acial)]TJ 0 -11.955 Td [(e)15(xpressions,)-303(head)-292(mo)15(v)15(ements,)-303(and)-293(body)-292(postures\))-292([)]TJ
0 0.44315 0.73724 rg 0 0.44315 0.73724 RG
[(10)]TJ
0 g 0 G
[(,)]TJ
0 0.44315 0.73724 rg 0 0.44315 0.73724 RG
[-293(37)]TJ
0 g 0 G
[(])]TJ 0 -11.955 Td [(with)-262(its)-262(o)25(wn)-261(grammar)-262(and)-262(le)15(xicon)-262([)]TJ
0 0.44315 0.73724 rg 0 0.44315 0.73724 RG
[(41)]TJ
0 g 0 G
[(].)-345(V)60(ision-based)-262(Con-)]TJ 0 -11.955 Td [(tinuous)-316(Sign)-317(Language)-316(Recognition)-316(\(CSLR\))-316(aims)-316(to)-317(auto-)]TJ 0 -11.955 Td [(matically)-357(recognize)-356(signs)-357(from)-357(image)-357(streams,)-383(which)-357(can)]TJ 0 -11.956 Td [(bridge)-288(the)-287(communication)-288(g)5(ap)-287(between)-288(the)-288(Deaf)-287(and)-288(hear)20(-)]TJ 0 -11.955 Td [(ing)-317(people.)-510(It)-316(also)-317(pro)15(vides)-316(more)-317(non-intrusi)25(v)15(e)-317(communi-)]TJ 0 -11.955 Td [(cation)-250(channel)-250(for)-250(sign)-250(language)-250(users.)]TJ 11.955 -16.385 Td [(Dif)25(ferent)-421(from)-422(speech)-421(recognition,)-465(the)-421(data)-422(collection)]TJ -11.955 -11.955 Td [(and)-335(annotation)-334(of)-335(sign)-334(language)-335(are)-335(costly)65(,)-355(which)-335(poses)-335(a)]TJ
0 g 0 G
0 g 0 G
0 g 0 G
0 g 0 G
0 g 0 G
ET
1 0 0 1 332.487 441.881 cm
q
.39682 0 0 .39682 0 0 cm
q
1 0 0 1 0 0 cm
/Im1 Do
Q
Q
1 0 0 1 -332.487 -441.881 cm
BT
/F66 8.9664 Tf 308.862 433.911 Td [(Figure)-266(1.)-266(Ov)15(ervie)25(w)-266(of)-266(the)-266(proposed)-266(non-iterati)25(v)15(e)-266(CLSR)-265(approach)]TJ 0 -10.959 Td [(with)-438(the)-438(visual)-438(alignment)-439(constrai)1(nt.)-875(T)80(o)-438(solv)15(e)-438(the)-438(insuf)25(\002cient)]TJ 0 -10.959 Td [(training)-325(of)-326(the)-325(feature)-325(e)15(xtractor)40(,)-344(the)-326(proposed)-325(V)135(A)40(C)-325(enhances)-325(the)]TJ 0 -10.959 Td [(generalization)-372(ability)-373(of)-372(the)-372(visual)-373(e)15(xtrac)1(tor)-373(by)-372(constraining)-372(the)]TJ 0 -10.959 Td [(feature)-250(space)-250(with)-250(the)-250(alignment)-250(supervision.)]TJ
0 g 0 G
/F66 9.9626 Tf 0 -22.009 Td [(signi\002cant)-252(problem)-252(for)-252(recognition)-252([)]TJ
0 0.44315 0.73724 rg 0 0.44315 0.73724 RG
[(2)]TJ
0 g 0 G
[(].)-317(Therefore,)-253(most)-252(re-)]TJ 0 -11.955 Td [(cent)-206(CSLR)-205(w)10(orks)-206(solv)15(e)-206(this)-206(problem)-205(in)-206(a)-206(weakly)-206(supervised)]TJ 0 -11.955 Td [(manner)-366(and)-366(adopt)-366(netw)10(ork)-366(architectures)-366(composed)-367(of)-366(the)]TJ 0 -11.955 Td [(feature)-227(e)15(xtractor)-227(and)-226(the)-227(alignment)-227(module.)-302(The)-227(feature)-227(e)15(x-)]TJ 0 -11.955 Td [(tractor)-380(abstracts)-380(visual)-380(information)-380(from)-379(each)-380(frame,)-413(and)]TJ 0 -11.956 Td [(the)-277(alignment)-277(module)-277(searches)-277(the)-277(possible)-277(alignments)-277(be-)]TJ 0 -11.955 Td [(tween)-299(visual)-298(features)-299(and)-298(the)-299(corresponding)-298(labeling.)-456(Dif-)]TJ 0 -11.955 Td [(ferent)-334(to)-334(those)-333(w)10(orks)-334([)]TJ
0 0.44315 0.73724 rg 0 0.44315 0.73724 RG
[(27)]TJ
0 g 0 G
[(,)]TJ
0 0.44315 0.73724 rg 0 0.44315 0.73724 RG
[-334(29)]TJ
0 g 0 G
[(,)]TJ
0 0.44315 0.73724 rg 0 0.44315 0.73724 RG
[-334(31)]TJ
0 g 0 G
[(])-333(adopt)-334(HMMs)-334(to)-334(update)]TJ 0 -11.955 Td [(frame-wise)-325(state)-325(labels)-325(for)-324(the)-325(feature)-325(e)15(xtractor)40(,)-344(Gra)20(v)15(es)]TJ/F70 9.9626 Tf 229.057 0 Td [(et)]TJ -229.057 -11.955 Td [(al)]TJ/F66 9.9626 Tf 7.761 0 Td [(.)-299([)]TJ
0 0.44315 0.73724 rg 0 0.44315 0.73724 RG
[(15)]TJ
0 g 0 G
[(])-299(pro)15(vide)-299(a)-298(more)-299(ele)15(g)5(ant)-299(solution)-299(so-called)-299(Connec-)]TJ -7.761 -11.955 Td [(tionist)-358(T)70(emporal)-358(Classi\002cation)-358(\(CTC\))-358(to)-358(align)-359(the)-358(predic-)]TJ 0 -11.956 Td [(tion)-450(and)-451(labeling)-450(by)-450(maximizing)-451(the)-450(sum)-450(of)-451(probability)]TJ 0 -11.955 Td [(of)-370(all)-370(feasible)-370(alignments,)-400(which)-370(is)-371(adopte)1(d)-371(by)-370(follo)25(wing)]TJ 0 -11.955 Td [(w)10(orks)-250([)]TJ
0 0.44315 0.73724 rg 0 0.44315 0.73724 RG
[(3)]TJ
0 g 0 G
[(,)]TJ
0 0.44315 0.73724 rg 0 0.44315 0.73724 RG
[-250(6)]TJ
0 g 0 G
[(,)]TJ
0 0.44315 0.73724 rg 0 0.44315 0.73724 RG
[-250(8)]TJ
0 g 0 G
[(,)]TJ
0 0.44315 0.73724 rg 0 0.44315 0.73724 RG
[-250(9)]TJ
0 g 0 G
[(,)]TJ
0 0.44315 0.73724 rg 0 0.44315 0.73724 RG
[-250(27)]TJ
0 g 0 G
[(,)]TJ
0 0.44315 0.73724 rg 0 0.44315 0.73724 RG
[-250(36)]TJ
0 g 0 G
[(,)]TJ
0 0.44315 0.73724 rg 0 0.44315 0.73724 RG
[-250(46)]TJ
0 g 0 G
[(].)]TJ 11.955 -12.026 Td [(Although)-517(CTC-based)-517(CSLR)-517(methods)-517(pro)15(vide)-517(con)40(v)15(e-)]TJ -11.955 -11.955 Td [(nience)-330(in)-330(training,)-350(pre)25(vious)-329(studies)-330([)]TJ
0 0.44315 0.73724 rg 0 0.44315 0.73724 RG
[(9)]TJ
0 g 0 G
[(,)]TJ
0 0.44315 0.73724 rg 0 0.44315 0.73724 RG
[-330(39)]TJ
0 g 0 G
[(])-330(sho)25(w)-330(that)-330(end-)]TJ 0 -11.956 Td [(to-end)-348(training)-347(limits)-348(the)-348(discriminati)25(v)15(e)-348(po)25(wer)-347(of)-348(the)-348(fea-)]TJ 0 -11.955 Td [(ture)-200(e)15(xtractor)55(.)-293(The)15(y)-200(le)25(v)15(erage)-199(the)-200(iterati)25(v)15(e)-199(training)-200(scheme)-200(to)]TJ 0 -11.955 Td [(enhance)-268(the)-267(feature)-268(e)15(xtractor)40(,)-272(which)-267(signi\002cantly)-268(impro)15(v)15(es)]TJ 0 -11.955 Td [(the)-436(performance.)-867(Ne)25(v)15(ertheless,)-482(it)-436(requires)-436(an)-436(additional)]TJ 0 -11.955 Td [(\002ne-tuning)-309(process)-309(besides)-308(the)-309(end-to-end)-309(training)-309(and)-309(in-)]TJ 0 -11.955 Td [(creases)-322(the)-322(training)-322(time.)-526(Se)25(v)15(eral)-322(recent)-322(w)10(orks)-323([)]TJ
0 0.44315 0.73724 rg 0 0.44315 0.73724 RG
[(6)]TJ
0 g 0 G
[(,)]TJ
0 0.44315 0.73724 rg 0 0.44315 0.73724 RG
[-322(36)]TJ
0 g 0 G
[(])-322(try)]TJ 0 -11.956 Td [(to)-285(accelerate)-284(this)-285(training)-285(scheme)-285(by)-284(adopting)-285(fully)-285(con)40(v)20(o-)]TJ 0 -11.955 Td [(lutional)-250(netw)10(orks)-250(and)-250(\002ne-grained)-250(labels.)]TJ 11.955 -12.026 Td [(In)-279(this)-280(study)65(,)-286(we)-279(re)25(visit)-280(CTC-based)-279(CSLR)-279(model)-279(at)-280(dif-)]TJ
0 g 0 G
0 g 0 G
ET
endstream
endobj
34 0 obj
<< /Length 46 >>
stream
q
1 0 0 1 0 0 cm
/60KXLjiGC7t_YGSlMvGQaw Do
Q
endstream
endobj
35 0 obj
<< /Length 46 >>
stream
q
1 0 0 1 0 0 cm
/e-gjPB2i4Okj2K0JtRTXjA Do
Q
endstream
endobj
36 0 obj
<< /CS /DeviceRGB /S /Transparency /Type /Group >>
endobj
37 0 obj
<< /Font << /F61 65 0 R /F65 66 0 R /F66 67 0 R /F68 68 0 R /F70 69 0 R >> /ProcSet [ /PDF /Text ] /XObject << /60KXLjiGC7t_YGSlMvGQaw 70 0 R /Im1 71 0 R /e-gjPB2i4Okj2K0JtRTXjA 72 0 R >> >>
endobj
38 0 obj
<< /Annots [ 73 0 R 74 0 R 75 0 R 76 0 R 77 0 R 78 0 R 79 0 R 80 0 R 81 0 R 82 0 R 83 0 R 84 0 R 85 0 R 86 0 R 87 0 R 88 0 R 89 0 R 90 0 R 91 0 R 92 0 R 93 0 R 94 0 R 95 0 R 96 0 R 97 0 R 98 0 R 99 0 R 100 0 R 101 0 R 102 0 R 103 0 R 104 0 R 105 0 R 106 0 R 107 0 R 108 0 R 109 0 R 110 0 R 111 0 R 112 0 R 113 0 R 114 0 R 115 0 R ] /Contents [ 116 0 R 117 0 R ] /MediaBox [ 0 0 612 792 ] /Parent 9 0 R /Resources 118 0 R /Type /Page >>
endobj
39 0 obj
<< /Annots [ 119 0 R 120 0 R 121 0 R 122 0 R 123 0 R 124 0 R 125 0 R ] /Contents [ 126 0 R 127 0 R ] /Group 128 0 R /MediaBox [ 0 0 612 792 ] /Parent 9 0 R /Resources 129 0 R /Type /Page >>
endobj
40 0 obj
<< /Annots [ 130 0 R 131 0 R 132 0 R 133 0 R 134 0 R 135 0 R 136 0 R 137 0 R 138 0 R 139 0 R 140 0 R 141 0 R ] /Contents [ 142 0 R 143 0 R ] /Group 144 0 R /MediaBox [ 0 0 612 792 ] /Parent 9 0 R /Resources 145 0 R /Type /Page >>
endobj
41 0 obj
<< /Annots [ 146 0 R 147 0 R 148 0 R 149 0 R 150 0 R ] /Contents [ 151 0 R 152 0 R ] /Group 153 0 R /MediaBox [ 0 0 612 792 ] /Parent 9 0 R /Resources 154 0 R /Type /Page >>
endobj
42 0 obj
<< /Annots [ 155 0 R 156 0 R 157 0 R 158 0 R 159 0 R 160 0 R 161 0 R 162 0 R 163 0 R 164 0 R 165 0 R ] /Contents [ 166 0 R 167 0 R ] /MediaBox [ 0 0 612 792 ] /Parent 9 0 R /Resources 168 0 R /Type /Page >>
endobj
43 0 obj
<< /Annots [ 169 0 R 170 0 R 171 0 R 172 0 R 173 0 R 174 0 R 175 0 R 176 0 R ] /Contents [ 177 0 R 178 0 R ] /Group 179 0 R /MediaBox [ 0 0 612 792 ] /Parent 10 0 R /Resources 180 0 R /Type /Page >>
endobj
44 0 obj
<< /Annots [ 181 0 R 182 0 R 183 0 R 184 0 R 185 0 R 186 0 R 187 0 R 188 0 R 189 0 R 190 0 R 191 0 R 192 0 R 193 0 R 194 0 R 195 0 R 196 0 R 197 0 R 198 0 R 199 0 R 200 0 R 201 0 R 202 0 R 203 0 R 204 0 R 205 0 R 206 0 R 207 0 R 208 0 R 209 0 R 210 0 R 211 0 R 212 0 R 213 0 R 214 0 R 215 0 R ] /Contents [ 216 0 R 217 0 R ] /MediaBox [ 0 0 612 792 ] /Parent 10 0 R /Resources 218 0 R /Type /Page >>
endobj
45 0 obj
<< /Annots [ 219 0 R 220 0 R 221 0 R 222 0 R 223 0 R 224 0 R 225 0 R 226 0 R 227 0 R 228 0 R 229 0 R 230 0 R 231 0 R 232 0 R 233 0 R 234 0 R 235 0 R 236 0 R 237 0 R 238 0 R 239 0 R 240 0 R 241 0 R 242 0 R 243 0 R 244 0 R 245 0 R 246 0 R 247 0 R 248 0 R 249 0 R 250 0 R 251 0 R 252 0 R 253 0 R 254 0 R 255 0 R 256 0 R 257 0 R 258 0 R 259 0 R 260 0 R 261 0 R 262 0 R 263 0 R ] /Contents [ 264 0 R 265 0 R ] /MediaBox [ 0 0 612 792 ] /Parent 10 0 R /Resources 266 0 R /Type /Page >>
endobj
46 0 obj
<< /Annots [ 267 0 R 268 0 R 269 0 R 270 0 R 271 0 R 272 0 R 273 0 R 274 0 R 275 0 R 276 0 R 277 0 R 278 0 R 279 0 R 280 0 R 281 0 R 282 0 R 283 0 R 284 0 R 285 0 R 286 0 R 287 0 R 288 0 R 289 0 R 290 0 R 291 0 R 292 0 R 293 0 R 294 0 R 295 0 R 296 0 R 297 0 R 298 0 R 299 0 R 300 0 R 301 0 R 302 0 R 303 0 R 304 0 R 305 0 R ] /Contents [ 306 0 R 307 0 R ] /MediaBox [ 0 0 612 792 ] /Parent 10 0 R /Resources 308 0 R /Type /Page >>
endobj
47 0 obj
<< /Limits [ (Doc-Start) (cite.camgoz2017subunets) ] /Names [ (Doc-Start) 309 0 R (Hfootnote.1) 310 0 R (cite.NIPS2017_900c563b) 311 0 R (cite.bowman2016generating) 312 0 R (cite.bragg2019sign) 313 0 R (cite.camgoz2017subunets) 314 0 R ] >>
endobj
48 0 obj
<< /Limits [ (cite.camgoz2020sign) (cite.cui2019deep) ] /Names [ (cite.camgoz2020sign) 315 0 R (cite.caruana1997multitask) 316 0 R (cite.cheng2020fully) 317 0 R (cite.cihan2018neural) 318 0 R (cite.cui2017recurrent) 319 0 R (cite.cui2019deep) 320 0 R ] >>
endobj
49 0 obj
<< /Limits [ (cite.distillation) (cite.graves2008novel) ] /Names [ (cite.distillation) 321 0 R (cite.dreuw2007speech) 322 0 R (cite.freeman1995orientation) 323 0 R (cite.gao2004chinese) 324 0 R (cite.graves2006connectionist) 325 0 R (cite.graves2008novel) 326 0 R ] >>
endobj
50 0 obj
<< /Limits [ (cite.graves2012supervised) (cite.hochreiter1997long) ] /Names [ (cite.graves2012supervised) 327 0 R (cite.graves2013speech) 328 0 R (cite.han2009modelling) 329 0 R (cite.hannun2014deep) 330 0 R (cite.he2016deep) 331 0 R (cite.hochreiter1997long) 332 0 R ] >>
endobj
51 0 obj
<< /Limits [ (cite.huang2018video) (cite.koller2016deephand) ] /Names [ (cite.huang2018video) 333 0 R (cite.ioffe2015batch) 334 0 R (cite.kim2017joint) 335 0 R (cite.koller2015continuous) 336 0 R (cite.koller2016deep) 337 0 R (cite.koller2016deephand) 338 0 R ] >>
endobj
52 0 obj
<< /Limits [ (cite.koller2017re) (cite.li2020word) ] /Names [ (cite.koller2017re) 339 0 R (cite.koller2019weakly) 340 0 R (cite.li2020overcoming) 341 0 R (cite.li2020reinterpreting) 342 0 R (cite.li2020transferring) 343 0 R (cite.li2020word) 344 0 R ] >>
endobj
53 0 obj
<< /Limits [ (cite.msasl) (cite.sanabria2018hierarchical) ] /Names [ (cite.msasl) 345 0 R (cite.niustochastic) 346 0 R (cite.ong2005automatic) 347 0 R (cite.pu2019iterative) 348 0 R (cite.pu2020boosting) 349 0 R (cite.sanabria2018hierarchical) 350 0 R ] >>
endobj
54 0 obj
<< /Limits [ (cite.sandler2006sign) (cite.wang2018connectionist) ] /Names [ (cite.sandler2006sign) 351 0 R (cite.simonyan2014very) 352 0 R (cite.sun2013discriminative) 353 0 R (cite.szegedy2015going) 354 0 R (cite.wang2017normface) 355 0 R (cite.wang2018connectionist) 356 0 R ] >>
endobj
55 0 obj
<< /Limits [ (cite.yang2019sf) (equation.4.4) ] /Names [ (cite.yang2019sf) 357 0 R (cite.zhou2020spatial) 358 0 R (equation.3.1) 359 0 R (equation.3.2) 360 0 R (equation.3.3) 361 0 R (equation.4.4) 362 0 R ] >>
endobj
56 0 obj
<< /Limits [ (equation.4.5) (figure.3) ] /Names [ (equation.4.5) 363 0 R (equation.4.6) 364 0 R (equation.4.7) 365 0 R (figure.1) 366 0 R (figure.2) 367 0 R (figure.3) 368 0 R ] >>
endobj
57 0 obj
<< /Limits [ (figure.4) (page.2) ] /Names [ (figure.4) 369 0 R (figure.5) 370 0 R (figure.6) 371 0 R (page.1) 372 0 R (page.10) 373 0 R (page.2) 374 0 R ] >>
endobj
58 0 obj
<< /Limits [ (page.3) (page.8) ] /Names [ (page.3) 375 0 R (page.4) 376 0 R (page.5) 377 0 R (page.6) 378 0 R (page.7) 379 0 R (page.8) 380 0 R ] >>
endobj
59 0 obj
<< /Limits [ (page.9) (section.4) ] /Names [ (page.9) 381 0 R (section*.1) 382 0 R (section.1) 383 0 R (section.2) 384 0 R (section.3) 385 0 R (section.4) 386 0 R ] >>
endobj
60 0 obj
<< /Limits [ (section.5) (subfigure.6.1) ] /Names [ (section.5) 387 0 R (section.6) 388 0 R (subfigure.3.1) 389 0 R (subfigure.3.2) 390 0 R (subfigure.3.3) 391 0 R (subfigure.6.1) 392 0 R ] >>
endobj
61 0 obj
<< /Limits [ (subfigure.6.2) (subsection.3.3) ] /Names [ (subfigure.6.2) 393 0 R (subsection.2.1) 394 0 R (subsection.2.2) 395 0 R (subsection.3.1) 396 0 R (subsection.3.2) 397 0 R (subsection.3.3) 398 0 R ] >>
endobj
62 0 obj
<< /Limits [ (subsection.4.1) (subsection.5.4) ] /Names [ (subsection.4.1) 399 0 R (subsection.4.2) 400 0 R (subsection.5.1) 401 0 R (subsection.5.2) 402 0 R (subsection.5.3) 403 0 R (subsection.5.4) 404 0 R ] >>
endobj
63 0 obj
<< /Limits [ (subsection.5.5) (table.5) ] /Names [ (subsection.5.5) 405 0 R (table.1) 406 0 R (table.2) 407 0 R (table.3) 408 0 R (table.4) 409 0 R (table.5) 410 0 R ] >>
endobj
64 0 obj
<< /Limits [ (table.6) (table.6) ] /Names [ (table.6) 411 0 R ] >>
endobj
65 0 obj
<< /BaseFont /ZUSOWF+CMSY9 /FirstChar 0 /FontDescriptor 412 0 R /LastChar 103 /Subtype /Type1 /ToUnicode 413 0 R /Type /Font /Widths 414 0 R >>
endobj
66 0 obj
<< /BaseFont /BCLGSQ+NimbusRomNo9L-Medi /Encoding 415 0 R /FirstChar 2 /FontDescriptor 416 0 R /LastChar 173 /Subtype /Type1 /ToUnicode 417 0 R /Type /Font /Widths 418 0 R >>
endobj
67 0 obj
<< /BaseFont /LOSSFU+NimbusRomNo9L-Regu /Encoding 415 0 R /FirstChar 2 /FontDescriptor 419 0 R /LastChar 180 /Subtype /Type1 /ToUnicode 420 0 R /Type /Font /Widths 421 0 R >>
endobj
68 0 obj
<< /BaseFont /TACSHG+NimbusMonL-Regu /Encoding 415 0 R /FirstChar 44 /FontDescriptor 422 0 R /LastChar 121 /Subtype /Type1 /ToUnicode 423 0 R /Type /Font /Widths 424 0 R >>
endobj
69 0 obj
<< /BaseFont /ILSYGH+NimbusRomNo9L-ReguItal /Encoding 415 0 R /FirstChar 2 /FontDescriptor 425 0 R /LastChar 122 /Subtype /Type1 /ToUnicode 426 0 R /Type /Font /Widths 427 0 R >>
endobj
70 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 428 0 R >> >> /Subtype /Form /Type /XObject /Length 382 /Filter /FlateDecode >>
stream
xe[n!Y7P|=Q'y(Ji;#$v0U-V\oZWɽ:5db,6IR6@6Pc"3K+GQ؎3T<9j`UT@
v5ڳtHN;9+7[^S؇:[E"8+sW\qEeКu6Bhsj;)y;Tp0nؚgʦFǃS-="z7ۺ9*&Q~ZPuí}C9嬂8l?Jܪmx%e8w](Rlq鞋t×e:=p=y0`XeJCWxau~uof{`r
endstream
endobj
71 0 obj
<< /BBox [ 0 0 476.28 238.08 ] /Filter /FlateDecode /FormType 1 /Group 36 0 R /PTEX.FileName (./imgs/title_figure.pdf) /PTEX.InfoDict 429 0 R /PTEX.PageNumber 1 /Resources << /ExtGState << /GS5 430 0 R /GS6 431 0 R >> /Font << /F1 432 0 R /F2 433 0 R /F3 434 0 R >> /ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ] /XObject << /Image10 435 0 R /Image11 436 0 R /Image9 437 0 R >> >> /Subtype /Form /Type /XObject /Length 3691 >>
stream
x[[o~ށ \m<b=YA͐Jr
(qr>?g\鳛w?^bf4)8R_gOz7fBٱ
_߆O?}͋?g
?O&
@hft,4PI)Ce!3o|OɎ)vl.@cJnhмu Į3nͶK_(/o1XoY
}?aacr?2l!
2-~DpZK^tq=aӿ^\O9%fM2%"+ԗv8'3@61
xb<{ws|B_a1>^WKVucG7?x[;z9?{iCWkwObwN5̸zV'ٷ >3hDȡq.xRhG0q!2)j5 Id?zt'Nfħ1!#%ay#ƑxѓH.Ȓ.(,p}ؘȒe Hd181bz?Y~b`-d] eomXIHGIL&d,!a4'l&VD$yOaU}K#g|Gev'fKKkN
oUEUU``i
Rr FU Zopc6`3l YtBh#h(Hv,ȴ@&arE͑D'Bc7ď
Y
º>DL⾼'|n,>sBeLTj¿rMp%˳ m
'sB9~/D+@A<|8QI@ ,H)W\[Ny,uX''K;[TH QmcVl t|]:D++#OCBƆtc+2#\'@$KiDhY3(O9
m~
sCxr0HiK!Ts@K99COfۚ8ܑܱ
)g"@5kt뫋7ew[KCE)10+Hli<IBIc8"X9F.92hCJb('Qb !/&57xFZxnL%<^rbj yO3)9u)P+cRp(l2X]غ*0\mFւlF~]uCD>֖$`U<`떀ة|wX