%PDF-1.3 1 0 obj << /Kids [ 3 0 R 4 0 R 5 0 R 6 0 R 7 0 R 8 0 R 9 0 R 10 0 R ] /Type /Pages /Count 8 >> endobj 2 0 obj << /Title (Attention\055Based Natural Language Person Retrieval) /Producer (PyPDF2) /Author (Tao Zhou\054 Muhao Chen\054 Jie Yu\054 Demetri Terzopoulos) /Subject (2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops) >> endobj 3 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 12 0 R /Resources << /XObject << /x8 13 0 R /x6 16 0 R /x12 19 0 R /x10 22 0 R >> /ExtGState << /s9 25 0 R /s11 28 0 R /a0 << /CA 1 /ca 1 >> /R14 31 0 R /s5 32 0 R /s7 35 0 R >> /Font << /F2 38 0 R /F1 39 0 R /R15 40 0 R /R17 44 0 R /R21 48 0 R /R19 52 0 R /R23 55 0 R /R25 58 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 62 0 R 63 0 R 64 0 R 65 0 R ] >> endobj 4 0 obj << /Parent 1 0 R /Rotate 0 /Contents 66 0 R /Resources << /XObject << /R56 67 0 R /R57 68 0 R >> /ExtGState << /R14 31 0 R >> /Font << /F2 69 0 R /F1 70 0 R /R15 40 0 R /R17 44 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /Group 71 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 72 0 R 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 ] /Type /Page >> endobj 5 0 obj << /Parent 1 0 R /Rotate 0 /Contents 95 0 R /Resources << /XObject << /R82 96 0 R /R81 97 0 R /R80 98 0 R /R79 99 0 R >> /ExtGState << /R14 31 0 R >> /Font << /R73 100 0 R /R77 103 0 R /R75 107 0 R /F2 111 0 R /F1 112 0 R /R15 40 0 R /R17 44 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /Group 71 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 113 0 R 114 0 R 115 0 R 116 0 R 117 0 R 118 0 R 119 0 R 120 0 R 121 0 R ] /Type /Page >> endobj 6 0 obj << /Parent 1 0 R /Rotate 0 /Contents 122 0 R /Resources << /XObject << /R96 123 0 R /R97 124 0 R >> /ExtGState << /R14 31 0 R >> /Font << /F2 125 0 R /R100 126 0 R /R17 44 0 R /R77 103 0 R /R75 107 0 R /R98 131 0 R /F1 136 0 R /R15 40 0 R /R73 100 0 R /R21 48 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /Group 71 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 137 0 R 138 0 R 139 0 R 140 0 R 141 0 R 142 0 R ] /Type /Page >> endobj 7 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 143 0 R /Resources << /ExtGState << /R14 31 0 R >> /Font << /F2 144 0 R /R73 100 0 R /R112 145 0 R /R100 126 0 R /R110 148 0 R /R116 152 0 R /R77 103 0 R /R114 157 0 R /R75 107 0 R /R98 131 0 R /F1 160 0 R /R15 40 0 R /R17 44 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 161 0 R 162 0 R 163 0 R ] >> endobj 8 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 164 0 R /Resources << /XObject << /R136 165 0 R >> /ExtGState << /R14 31 0 R >> /Font << /R73 100 0 R /R17 44 0 R /R75 107 0 R /F2 166 0 R /F1 167 0 R /R15 40 0 R /R100 126 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 168 0 R 169 0 R 170 0 R 171 0 R 172 0 R 173 0 R 174 0 R 175 0 R 176 0 R 177 0 R 178 0 R 179 0 R 180 0 R ] >> endobj 9 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 181 0 R /Resources << /XObject << /R144 182 0 R /R143 183 0 R /R142 184 0 R >> /ExtGState << /R14 31 0 R >> /Font << /F2 185 0 R /F1 186 0 R /R25 58 0 R /R15 40 0 R /R17 44 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /Group 71 0 R /MediaBox [ 0 0 612 792 ] /Annots [ ] >> endobj 10 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 187 0 R /Resources << /ExtGState << /R14 31 0 R >> /Font << /F2 188 0 R /F1 189 0 R /R25 58 0 R /R17 44 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ ] >> endobj 11 0 obj << /Type /Catalog /Pages 1 0 R >> endobj 12 0 obj << /Length 18622 >> stream q q q 0.10000 0 0 0.10000 0 0 cm /R14 gs 0 g q 10 0 0 10 0 0 cm BT /R15 14.34620 Tf 1 0 0 1 138.14800 675.06700 Tm [ (Attention\055based) -249.99500 (Natural) -249.99300 (Language) -250.01000 (P) 20.00610 (erson) -250.00200 (Retrie) 14.98930 (v) 9.99625 (al) ] TJ /R17 11.95520 Tf 21.73280 -37.85820 Td [ (T) 80.01370 (ao) -250.00900 (Zhou) ] TJ /R19 7.97010 Tf 45.86020 4.33906 Td [ (1) -0.30019 ] TJ /R17 11.95520 Tf 4.73203 -4.33906 Td [ (\054) -250.01200 (Muhao) -249.99300 (Chen) ] TJ /R19 7.97010 Tf 68.07300 4.33906 Td [ (1) -0.30019 ] TJ /R17 11.95520 Tf 4.73203 -4.33906 Td [ (\054) -250.01200 (Jie) -250.00600 (Y) 110.99600 (u) ] TJ /R19 7.97010 Tf 35.53090 4.33906 Td [ (2) -0.30019 ] TJ /R17 11.95520 Tf 4.73203 -4.33906 Td [ (\054) -250.01200 (Demetri) -250.00200 (T) 70.01640 (erzopoulos) ] TJ /R19 7.97010 Tf 107.06900 4.33906 Td [ (1) -0.30019 ] TJ /R17 11.95520 Tf -225.60700 -18.28590 Td [ (Uni) 24.99570 (v) 14.98510 (ersity) -249.98900 (of) -250.01400 (California\054) -249.99300 (Los) -249.98900 (Angeles) ] TJ /R19 7.97010 Tf 180.48500 4.33789 Td [ (1) -0.30019 ] TJ /R17 11.95520 Tf -171.94900 -18.28590 Td [ (SAIC) -249.99300 (Inno) 15.01580 (v) 24.98340 (ation) -250.00200 (Center) 39.98490 (\054) -250.01200 (San) -249.98900 (Jose) ] TJ /R19 7.97010 Tf 163.41400 4.33789 Td [ (2) -0.30019 ] TJ /R21 8.96640 Tf -232.36000 -18.28590 Td [ (f) -0.89990 ] TJ /R23 8.96640 Tf 4.60820 0 Td [ (taozhou\054) -600.02100 (muhaochen\054) -599.98800 (dt) ] TJ /R21 8.96640 Tf 118.35600 0 Td [ (g) -0.90126 ] TJ /R23 8.96640 Tf 4.60703 0 Td (\100cs\056ucla\056edu) Tj /R17 11.95520 Tf 64.55780 0 Td (\073) Tj /R23 8.96640 Tf 6.31211 0 Td (jerry\056j\056yu\100gmail\056com) Tj /R15 11.95520 Tf -197.03900 -41.04610 Td (Abstract) Tj /R25 9.96260 Tf -83.92770 -23.92890 Td [ (F) 105.00600 (ollowing) -205.00900 (the) -205.01700 (r) 37.01960 (ecent) -204.98700 (pr) 44.98390 (o) 10.00320 (gr) 36.98520 (ess) -205.01700 (in) -205.01200 (ima) 10.01300 (g) 10.00320 (e) -204.99000 <636c6173736902636174696f6e> -205.00700 (and) ] TJ -11.95510 -11.95510 Td [ (captioning) -344.01300 (using) -345.01200 (deep) -343.98100 (learning) 9.99466 (\054) -368.01500 (we) -344.99300 (de) 15.01710 (velop) -344.01400 (a) -344.01600 (no) 10.00810 (vel) -345.00400 (natu\055) ] TJ 11.95590 TL T* [ (r) 14.98400 (al) -275 (langua) 9.99159 (g) 10.00320 (e) -275.01700 (per) 10.00570 (son) -274.99700 (r) 37.01960 (etrie) 14.99010 (val) -275.00600 (system) -275 (based) -274.99600 (on) -275.98300 (an) -275.00300 (attention) ] TJ 11.95510 TL T* [ (mec) 15.01100 (hanism\056) -692.98800 (Mor) 36.98650 (e) -377.98100 <73706563690263616c6c79> 54.99690 (\054) -408.98800 (given) -378.00700 (the) -376.98900 (description) -378.01000 (of) -378.00300 (a) ] TJ T* [ (per) 10.00510 (son\054) -348.98200 (the) -328.99600 (goal) -330.00100 (is) -328.99000 (to) -328.99200 (localize) -329.99400 (the) -328.99700 (per) 10.00570 (son) -328.98700 (in) -328.99200 (an) -330.01400 (ima) 10.01300 (g) 10.00320 (e) 15.01220 (\056) -547.98500 (T) 92 (o) ] TJ T* [ (this) -252.99100 (end\054) -255.00800 (we) -253.00700 <0272> 10.00940 (st) -253.98300 (construct) -252.98400 (a) -253.01200 (benc) 15.01830 (hmark) -253.98200 (dataset) -252.98500 (for) -254.01900 (natur) 15.00730 (al) ] TJ T* [ (langua) 9.99159 (g) 10.00320 (e) -263.01900 (per) 10.00570 (son) -263 (r) 37.01960 (etrie) 14.99010 (val\056) -348.01100 (T) 92 (o) -263.01000 (do) -263.00500 (so\054) -265.98100 (we) -263.00500 (g) 10.00320 (ener) 15.01960 (ate) -263.00800 (bounding) ] TJ T* [ (boxes) -338.99500 (for) -338.00400 (per) 10.00570 (sons) -339.01900 (in) -338.01000 (a) -338.99700 (public) -337.99800 (ima) 10.01300 (g) 10.00320 (e) -339.00700 (dataset) -339.01000 (fr) 44.98510 (om) -338.01500 (the) -338.99500 (se) 39.99580 (g\055) ] TJ 11.95630 TL T* [ (mentation) -370.99600 (masks\054) -401.01300 (whic) 14.99870 (h) -370.99200 (ar) 36.98520 (e) -371.98200 (then) -370.98500 (annotated) -370.99700 (with) -371.00400 (descrip\055) ] TJ 11.95470 TL T* [ (tions) -224.01000 (and) -223.98700 (attrib) 19.99540 (utes) -223.98800 (using) -224.01300 (the) -223.99400 (Amazon) -222.99300 (Mec) 15.00980 (hanical) -224.01200 (T) 54.98590 (urk\056) -302.01800 (W) 91.98710 (e) ] TJ T* [ (then) -229.98800 (adopt) -230.01300 (a) -229.99600 (r) 37.01960 (e) 39.98840 (gion) -230.01800 (pr) 44.98390 (opos) 1.00106 (a) -1.01454 (l) -229.01300 (network) -230.01200 (in) -229.98900 (F) 75.01200 (aster) -229.98100 (R\055CNN) -229.99800 (as) -229.98900 (a) ] TJ T* [ (candidate) -303.99700 (r) 37.01960 (e) 39.98900 (gion) -304.98600 (g) 10.00320 (ener) 15.01960 (ator) 110.98500 (\056) -473.01700 (The) -304.01300 (cr) 45.01330 (opped) -304.99300 (ima) 10.01300 (g) 10.00320 (es) -303.98600 (based) -304.99100 (on) ] TJ T* [ (the) -213.01500 (r) 37.01960 (e) 39.98900 (gion) -213.98100 (pr) 44.98390 (oposals) -212.98400 (as) -213.01200 (well) -213 (as) -213.99200 (the) -213.01600 (whole) -212.99600 (ima) 10.01300 (g) 10.00320 (es) -212.98200 (with) -214.01100 (atten\055) ] TJ 11.95590 TL T* [ (tion) -205 (weights) -204.00500 (ar) 36.98520 (e) -204.99000 (fed) -203.99800 (into) -205 (Con) 40.01660 (volutional) -205.01500 (Neur) 14.99750 (al) -203.99300 (Networks) -204.98300 (for) ] TJ 11.95510 TL T* [ (visual) -191 (featur) 37 (e) -192.01200 (e) 19.99180 (xtr) 14.98650 (action\054) -202.99800 (while) -190.99900 (the) -192 (natur) 15.00730 (al) -191.01500 (langua) 9.99098 (g) 10.00320 (e) -192.01200 (e) 19.99180 (xpr) 36.98160 (es\055) ] TJ T* [ (sion) -351.01600 (and) -350.00500 (attrib) 19.99540 (utes) -350.98700 (ar) 36.98520 (e) -351.00500 (input) -351.01000 (to) -350.00800 (Bidir) 37.00360 (ectional) -351.01300 (Long) -351.01000 (Short\055) ] TJ T* [ (T) 91.99870 (erm) -307.00300 (Memory) -307.98000 (\050BLSTM) 1.01576 (\051) -307.99100 (models) -306.98600 (for) -306.98900 (te) 20 (xt) -307.00500 (featur) 37.00120 (e) -307.99300 (e) 19.99180 (xtr) 14.98770 (action\056) ] TJ T* [ (The) -324.98900 (visual) -323.99800 (and) -324.99000 (te) 20 (xt) -325.00200 (featur) 37 (es) -325.00300 (ar) 36.98520 (e) -323.99000 (inte) 40.00800 (gr) 14.99010 (ated) -324.99200 (to) -324.99200 (scor) 36.98650 (e) -325.00900 (r) 37.01830 (e) 39.98840 (gion) ] TJ T* [ (pr) 44.98390 (oposals\054) -380.01300 (and) -352.98500 (the) -354.01200 (one) -354.01300 (with) -353.98600 (the) -352.99300 (highest) -353.98700 (scor) 36.98650 (e) -353.98500 (is) -354.00700 (r) 37.01830 (etrie) 14.99010 (ved) ] TJ 11.95590 TL T* [ (as) -273.98200 (the) -273.98600 (output) -275.01800 (of) -273.98100 (our) -274.01700 (system\056) -383.01800 (The) -274.01800 (e) 19.99180 (xperimental) -274 (r) 37.01830 (esults) -274.98600 (show) ] TJ 11.95510 TL T* [ <7369676e690263616e74> -203.99600 (impr) 44.99370 (o) 10.00320 (vement) -204.98300 (o) 10.00320 (ver) -204.00800 (the) -203.99800 (state\055of\055the\055art) -203.99500 (method) -205.00500 (for) ] TJ T* [ (g) 10.00320 (eneric) -211.99800 (object) -213.00700 (r) 37.01960 (etrie) 14.99010 (val) -211.99600 (and) -213.00900 (this) -212.01700 (line) -211.98900 (of) -213.01100 (r) 37.01830 (esear) 36.98160 (c) 15.01220 (h) -211.99900 (pr) 44.98510 (omises) -213.00100 (to) ] TJ T* [ <62656e650274> -249.98400 (sear) 36.98710 (c) 15.01280 (h) -249.99300 (in) -249.98500 (surveillance) -250.00500 (video) -249.98900 (foota) 9.98853 (g) 10.00320 (e) 15.01220 (\056) ] TJ /R15 11.95520 Tf 35.92230 TL T* [ (1\056) -249.99000 (Intr) 18.01460 (oduction) ] TJ /R17 9.96260 Tf 11.95510 -18.94800 Td [ (V) 59.99310 (ideo) -235.98700 (surv) 14.99010 (eillance) -237.01100 (cameras) -236.00200 (ha) 19.99790 (v) 14.98280 (e) -236.98500 (been) -235.99700 (deplo) 10.01550 (yed) -235.99500 (in) -237.00700 (man) 14.99010 (y) ] TJ -11.95510 -11.95470 Td [ (places\227in) -252.98900 (stores) -252.01900 (and) -253.01000 (homes) -252.99200 (for) -251.99800 (indoor) -252.99200 (monitoring\054) -252.99700 (as) -253.01400 (well) ] TJ 11.95630 TL T* [ (as) -291.98900 (in) -292.01700 (roadw) 10.00380 (ays) -293.00300 (and) -291.98400 (parking) -292.01000 (lots) -292.00400 (for) -293.01100 (wide\055) 0.98758 (area) -292.98800 (observ) 24.99090 (ation\056) ] TJ 11.95510 TL T* [ (Due) -309.00400 (to) -308.99500 (the) -309 (ubiquity) -309 (of) -309.98600 (these) -308.99700 (de) 25.01670 (vices\054) -324.01200 (ena) 1.01699 (b) -1.01454 (l) 0.98758 (ing) -310.01000 (more) -308.99700 (ef) 25.00810 (fec\055) ] TJ T* [ (ti) 24.99030 (v) 14.98280 (e) -269.01800 (technologies) -269.01400 (for) -269.99500 (surv) 14.99010 (eillance) -269.00500 (data) -269.01000 (analysis) -268.98700 (has) -269.98700 (become) ] TJ T* [ (an) -345 (ur) 17.99880 (gent) -344.98800 (challenge\056) -594.00400 (This) -345.01800 (demands) -345.01600 (automatic) -345.00900 (recognition) ] TJ T* [ (and) -211.99800 (characteri) 1.01086 (zation) -211.98800 (of) -212.00300 (persons) -211.00400 (in) -211.99200 (the) -211.01600 (lar) 17.99700 (ge) -212.00400 (quantities) -210.98200 (of) -212.00100 (im\055) ] TJ T* [ (ages\057videos\054) -208.00800 (and) -196.98100 (also) -197.01200 (requires) -197.99800 (machines) -197.00400 (to) -197.01400 (mak) 10.01060 (e) -196.99200 (such) -197.99400 (visual) ] TJ 11.95590 TL T* [ (information) -239.99500 (of) -239.99800 (persons) -239 (compatible) -240.00300 (with) -240.01600 (human) -240.01600 (understand\055) ] TJ 11.95510 TL T* [ (ing\056) -756.01100 (Hence\054) -436.00800 (it) -399.01700 (is) -399.01800 (highly) -398.99700 (desirable) -399.01700 (for) -398.01400 (a) -398.99700 (system) -399 (to) -399.01900 (match) ] TJ T* [ (visual) -240.01200 (objects) -240.01600 (with) -240.01600 (corresponding) -240.00500 (human) -240.01600 (language) -239.98400 (descrip\055) ] TJ T* [ (tions\056) -736.98100 (The) -392.98900 (understanding) -392.00800 (of) -392.01200 (v) -1.01454 (i) 0.98758 (sual) -393.01900 (contents) -391.98600 (is) -393.02000 (e) 15.01220 (xtremely) ] TJ 258.75000 473.32000 Td [ (challenging) -256.98200 (due) -257.01100 (to) -257.00400 (f) 9.99588 (actors) -257.01100 (such) -257.98600 (as) -257.01600 (lo) 24.98600 (w) -257.01100 (resolution\054) -259.01100 (deforma\055) ] TJ T* [ (tion\054) -339.99500 (and) -323 (occlusion\054) -339.98700 (and) -323 (so) -322.01500 (is) -322.99000 (the) -322.01500 (understanding) -321.98100 (of) -323.00500 (human) ] TJ 11.95590 TL T* [ (language) -342.98700 (e) 15.01220 (xpressions) -343.00400 (where) -343.00700 (semantics) -343.01100 (are) -343.01100 (latent) -344.00100 (and) -342.99700 (hard) ] TJ 11.95510 TL T* [ (to) -249.98500 (quantify) 64.99650 (\056) ] TJ 11.95510 -15.48910 Td [ (Se) 24.98600 (v) 14.98280 (eral) -349.98600 (recent) -349.00500 (papers) -350.01500 (report) -348.99600 (promising) -350.01000 (progress) -348.98600 (in) -350.01000 (this) ] TJ -11.95510 -11.95510 Td [ (area\056) -905 (Feris) -449.00400 (et) -448.00100 (al\056) -449.01800 (\1338\135) -448.00600 (retrie) 25.01540 (v) 14.98280 (ed) -448.00400 (human) -448.99900 (f) 9.99343 (aces) -448.00900 (from) -449.01300 (video) ] TJ T* [ (streams) -385.01400 (gi) 24.98850 (v) 14.98280 (en) -384.99500 <636c61737369026564> -385.00200 (and) -384.00900 (weighted) -385.00900 (attrib) 20.01630 (utes\054) -419.01300 (b) 20.00160 (ut) -384.98500 (the) 14.98770 (y) ] TJ 11.95590 TL T* [ (could) -237.98200 (not) -237.98200 (handle) -237.98500 (conte) 14.98520 (xts) -237.98000 (lik) 10.01790 (e) -238.00400 (people) -237.98500 (and) -237.99500 (scenes) 0.98513 (\054) -241.01400 (as) -238 (well) -237.98500 (as) ] TJ 11.95510 TL T* [ (language) -215.98800 (inputs\056) -299.01100 (Moreo) 14.99260 (v) 14.98280 (er) 39.98350 (\054) -223.01700 (Socher) -215.98800 (et) -215.99800 (al\056) -215.99800 (\133) ] TJ /R15 9.96260 Tf 165.55500 0 Td (\077) Tj /R17 9.96260 Tf 4.98086 0 Td [ (\135) -216.00800 (e) 25.01050 (v) 24.98110 (aluated) -215.99300 (if) -215.99800 (the) ] TJ -170.53600 -11.95510 Td [ (image) -310.99200 (contains) -311.01900 (seen) -311.00400 (or) -311.00400 (unseen) -310.99200 (classes\054) -326.99000 (whereas) -311.00700 (Lazaridou) ] TJ T* [ (et) -269.99200 (al\056) -269.98900 (\13318\135) -269.99200 (focused) -271.00400 (on) -269.98400 (predicting) -270.00400 (k) 10.00320 (e) 15.01220 (yw) 10.00810 (ords) -269.97900 (for) -269.99400 (images\056) -371.01200 (The) ] TJ T* [ (common) -281.01700 (problem) -282.00400 (of) -281.00900 (these) -281.00200 (ef) 25.00560 (forts) -281.00200 (is) -282.01700 (that) -280.99700 (the) 14.98770 (y) -281.00700 (rely) -281.00700 (on) -281.98200 (tradi\055) ] TJ T* [ (tional) -222.01400 (language) -221.00700 (models) -222.01000 (that) -221.98700 (are) -222.01200 (not) -221.00200 (able) -221.99700 (to) -221.99200 (handle) -221.98700 (sentences) ] TJ 11.95590 TL T* [ (with) -306.98600 (rich) -307.98300 (semantics) -307.01500 (and) -307.00300 (conte) 14.98770 (xts\056) -482.99800 (W) 39.99330 (ith) -306.98800 (the) -306.99800 (resurrection) -308.01200 (of) ] TJ 11.95510 TL T* [ (neural) -271.00900 (netw) 10.00810 (orks\054) -276.01800 (deep) -269.99200 (learning) -270.99400 (has) -271.00900 (no) 24.98600 (w) -269.98900 (become) -271.00400 (popular) -270.99400 (in) ] TJ T* [ (visual) -401.98400 (recognition) -402.01600 (and) -402.00600 (natural) -401.99700 (language) -401.99700 (processing) -401.98200 (tasks\056) ] TJ T* [ (Ne) 25.01300 (v) 14.98280 (ertheless\054) -290.99300 (an) -283.99200 (attent) 0.99738 (i) 24.98360 (v) 14.98280 (e) -283.99700 (natural\055language\055dri) 25.00560 (v) 14.98280 (en) -283.01200 (retrie) 25.01540 (v) 24.98110 (al) ] TJ T* [ (that) -360.98400 (localizes) -360.00400 (the) -360.99100 (most) -359.98900 (w) 10.00320 (orthwhile) -361.00800 (objects\054) -387.98900 (including) -360.99400 (per) 19.99180 (\055) ] TJ 11.95630 TL T* [ (sons\054) -250.01200 (still) -250.00200 (remains) -250.01700 (lar) 17.99700 (gely) -249.98300 (unsupported\056) ] TJ 11.95510 -15.48910 Td [ (In) -301.98600 (this) -302.00400 (paper) 39.99090 (\054) -313.98400 (we) -301.99100 (address) -302.01800 (the) -300.99900 (follo) 24.99580 (wing) -302.00800 (problem\072) -414.01400 (Gi) 24.98600 (v) 14.98280 (en) ] TJ -11.95510 -11.95470 Td [ (an) -283.99200 (image) -282.99700 (of) -283.98900 (multiple) -283.00400 (persons) -284.01100 (and) -283.00700 (a) -283.99700 (sentence) -283.00700 (that) -284.01600 (describes) ] TJ 11.95510 TL T* [ (a) -355.98500 (person\054) -381.99800 (we) -355.98200 (w) 10 (ant) -354.99200 (to) -356.00700 (localize) -356.00900 (the) -356.00900 (most) -355.00900 (rele) 25.01050 (v) 24.98110 (ant) -356.00900 (person) -356.00400 (in) ] TJ T* [ (the) -406.00300 (image) -407.01500 (based) -406.00300 (on) -406.98100 (both) -405.98600 (the) -406.98100 (te) 14.98280 (xt) -406 (description) -406.99600 (and) -406.00600 (the) -406.98100 (vi\055) ] TJ 11.95590 TL T* [ (sual) -342.00700 (features\056) -586.99800 (F) 14.99260 (or) -342.01900 (e) 15.01220 (xample\054) -365.01300 (to) -342.01200 (retrie) 25.01540 (v) 14.98280 (e) -341.98700 (the) -342.01200 (rider) -343.00200 (\050red) -341.98700 (rect\055) ] TJ 11.95510 TL T* [ (angle) -306.02000 (in) -306.99500 (Figure) -306.00300 (1D\051) -307.00300 (gi) 24.98850 (v) 14.98280 (en) -305.98800 (a) -305.99300 (description) -307.01200 (saying) -306.00300 (\223an) -307.01200 (elderly) ] TJ T* [ (man) -279 (wearing) -279.00500 (a) -279.01700 (light) -280 (hat) 0.98758 (\054) -287.00800 (plaid) -278.99300 (shirt) -278.98800 (and) -279.00700 (dark) -279.01200 (pants) -280.01200 (riding) -278.98800 (a) ] TJ T* [ (bik) 10.01550 (e\224) -260.98600 (and) -261.01000 (attrib) 20.01630 (utes) -261 (such) -261.00300 (as) -261.01500 (\223male\224) -261.01500 (and) -261.01000 (\223on) -261.01000 (the) -261.00500 (right) -260.99600 (side\224\054) ] TJ T* [ (one) -297.98500 (must) -297.01900 (look) -298.00700 (at) -297.00700 (all) -298.01900 (the) -297.00200 (three) -297.99000 (persons) -296.99000 (to) -298.01900 (decide) -297.00400 (which) -298.01400 (one) ] TJ T* [ (best) -376.98100 (matches) -377.01800 (the) -376.98900 (description\056) -690.00300 (In) -376.99600 (f) 9.99343 (act\054) -408.99100 (a) -377 (similar) -377 (w) 10.00320 (ork) -376.99100 (has) ] TJ 11.95590 TL T* [ (been) -249.01500 (implemented) -249.00700 (in) -249.98500 (\13326\135) -249.01200 (by) -249.00700 (jointly) -249.01200 (processing) -248.98800 (images) -250.01700 (and) ] TJ 11.95510 TL T* [ (te) 14.98030 (xt) -370.00400 (mainly) -369.98500 (relying) -369.99500 (on) -370.00700 (pre\055trained) -370.01200 (objects) -371.01700 (and) -370.01200 (kno) 24.99090 (wledge) ] TJ T* [ (graphs\056) -572.00500 (In) -337.98000 (contrast) -336.98300 (to) -338.01000 (pre) 25.01050 (vious) -337.01200 (w) 10.00320 (orks\054) -358.98400 (we) -337.98300 (de) 25.01540 (v) 14.98280 (elop) -336.98800 (a) -337.98800 (deep) ] TJ T* [ (learning) -379.99800 (frame) 25.01300 (w) 10 (ork) -380.00800 (to) -380.98300 (learn) -380.01500 (a) -380.02000 (scoring) -379.99500 (function) -379.99100 (that) -381.02000 (tak) 10.00810 (es) ] TJ T* [ (the) -332.99600 (te) 14.98030 (xt) -332.99100 (query) 65.01610 (\054) -354.01700 (at) 0.98268 (trib) 20.01380 (utes) -332.98900 (and) -332.99800 (images) -332.98400 (as) -333.00300 (input\054) -353.98500 (and) -332.99800 (outputs) ] TJ 11.95590 TL T* [ (scores) -259.99100 (for) -259.99800 (the) -259.00800 (candidate) -259.98400 (re) 15.00730 (gions\056) -339.01200 (In) -259.99600 (practice\054) -261.99500 (this) -260.01100 (line) -259.98100 (of) -259.99600 (re\055) ] TJ 11.95510 TL T* [ (search) -252.98200 (moti) 25.00810 (v) 24.98110 (ates) -253.99200 (man) 14.99010 (y) -253.01200 (applications\054) -253.99200 (such) -253.98700 (as) -253.01700 (the) -253.00700 (retrie) 25.01540 (v) 24.98110 (al) -253.99200 (of) ] TJ T* [ (suspects) -224.01400 (and) -223.99700 (missing) -223.99200 (persons\054) -228.99600 (when) -223.99200 (the) -225.01100 (im) 1 (ages) -225.02100 (or) -224.00200 (videos) -224.01700 (of) ] TJ T* [ (the) -253.98900 (tar) 17.99700 (get) -255.00900 (person) -253.98200 (are) -254.98900 (hidden) -254.01200 (in) -253.98700 (the) -255.00600 (surv) 14.99260 (eillance) -253.98700 (data\056) -323 (In) -255.01600 (this) ] TJ T* [ (case\054) -281.99700 (the) -275.00600 (police) -276.01800 (can) -275.01500 (rely) -275.98800 (only) -274.98800 (on) -275.00300 (the) -275.98300 (witness\047) 54.99570 (s) -275.00800 (te) 14.98280 (xt) -275.98300 (descrip\055) ] TJ (tion\056) ' ET Q Q Q q q 1 1 1 rg /a0 gs 48.40600 786.42200 515.18800 -52.69900 re f q /s5 gs /x6 Do Q q /s7 gs /x8 Do Q q /s9 gs /x10 Do Q q /s11 gs /x12 Do Q Q Q Q q 1 0 0 1 0 0 cm BT /F1 12 Tf 14.40000 TL ET 1 1 1 rg n 270 32 72 14 re f* 0.50000 0.50000 0.50000 rg BT /F2 9 Tf 10.80000 TL ET BT 1 0 0 1 301.50000 35 Tm (27) Tj T* ET Q endstream endobj 13 0 obj << /Filter /FlateDecode /Resources << /ExtGState << /a0 << /CA 1 /ca 1 >> >> /XObject << /x18 14 0 R >> >> /Length 28 /Group << /Type /Group /S /Transparency /CS /DeviceRGB /I true >> /BBox [ 78 746 96 765 ] /Type /XObject /Subtype /Form >> stream x+O4PH/VЯ0Pp 0 endstream endobj 14 0 obj << /Filter /FlateDecode /Resources 15 0 R /Length 107 /Type /XObject /BBox [ 78 746 96 765 ] /Subtype /Form >> stream xe AC̬wʠ =p,?]%+H-
Jc "82w8VSnGW;"
endstream
endobj
15 0 obj
<<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
>>
endobj
16 0 obj
<<
/Filter /FlateDecode
/Resources <<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
/XObject <<
/x15 17 0 R
>>
>>
/Length 28
/Group <<
/Type /Group
/S /Transparency
/CS /DeviceRGB
/I true
>>
/BBox [ 67 752 84 775 ]
/Type /XObject
/Subtype /Form
>>
stream
x+O4PH/VЯ04Up
0
endstream
endobj
17 0 obj
<<
/Filter /FlateDecode
/Resources 18 0 R
/Length 228
/Type /XObject
/BBox [ 67 752 84 775 ]
/Subtype /Form
>>
stream
xeQKn!s ?FPav6R٪TS.
b];15YyR
{7QL.\:Rv/x9l+L7h%1!}i/AI(kz"U&,YO![R hg{3}4/GyYF:!w}Gn+'xJcO9i뽼_-:`
endstream
endobj
18 0 obj
<<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
>>
endobj
19 0 obj
<<
/Filter /FlateDecode
/Resources <<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
/XObject <<
/x24 20 0 R
>>
>>
/Length 28
/Group <<
/Type /Group
/S /Transparency
/CS /DeviceRGB
/I true
>>
/BBox [ 112 751 500 772 ]
/Type /XObject
/Subtype /Form
>>
stream
x+O4PH/VЯ02Qp
0
endstream
endobj
20 0 obj
<<
/Filter /FlateDecode
/Resources 21 0 R
/Length 56490
/Type /XObject
/BBox [ 112 751 500 772 ]
/Subtype /Form
>>
stream
xtIJ5*|^~x?P9)<<ҥ+kh4}/?>]漿վ]Z Ǭ57W'b~?Kwk6:>rAニxeG1k-/LfߗA̺(egk5>
=KO1{Vux}Ol|_s؟|*>5=k|*a5x}On :97~1s7rX {|>IQQo6|'\#x_ڽC}M7(췀(N
سW#^Ň}|zblnB'\o=r;$s:z-
V3+έ昵 WD/nl ;q"3Pg휅\3C`k7760c9-Cȸ