%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 11 0 R ] /Type /Pages /Count 9 >> endobj 2 0 obj << /Title (AMC\072 Attention guided Multi\055modal Correlation Learning for Image Search) /Producer (PyPDF2) /Author (Kan Chen\054 Trung Bui\054 Chen Fang\054 Zhaowen Wang\054 Ram Nevatia) /Subject (2017 IEEE Conference on Computer Vision and Pattern Recognition) >> endobj 3 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 13 0 R /Resources << /XObject << /x8 14 0 R /R37 17 0 R /x6 20 0 R /x12 23 0 R /x10 26 0 R /R40 29 0 R >> /ExtGState << /s9 32 0 R /s5 35 0 R /s7 38 0 R /R29 41 0 R /s11 42 0 R /R16 45 0 R /a0 << /CA 1 /ca 1 >> /R32 46 0 R >> /Font << /R41 47 0 R /F2 51 0 R /F1 52 0 R /R17 53 0 R /R21 57 0 R /R19 60 0 R /R23 64 0 R /R35 67 0 R /R25 71 0 R /R33 75 0 R /R27 79 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] /ColorSpace << /R39 30 0 R /R31 18 0 R >> >> /MediaBox [ 0 0 612 792 ] /Annots [ 83 0 R 84 0 R 85 0 R 86 0 R 87 0 R 88 0 R 89 0 R 90 0 R ] >> endobj 4 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 91 0 R /Resources << /ExtGState << /R16 45 0 R >> /Font << /F2 92 0 R /R19 60 0 R /F1 93 0 R /R27 79 0 R /R17 53 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 116 0 R 117 0 R 118 0 R 119 0 R 120 0 R 121 0 R 122 0 R ] >> endobj 5 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 123 0 R /Resources << /XObject << /R101 124 0 R /R100 125 0 R /R87 126 0 R /R96 127 0 R /R97 128 0 R >> /ExtGState << /R32 46 0 R /R94 129 0 R /R86 130 0 R >> /Font << /R98 131 0 R /R88 135 0 R /R122 139 0 R /R120 142 0 R /R90 146 0 R /R92 150 0 R /R108 154 0 R /R118 158 0 R /R112 163 0 R /R110 166 0 R /R102 171 0 R /R116 175 0 R /R104 178 0 R /R114 182 0 R /R106 185 0 R /F2 188 0 R /F1 189 0 R /R17 53 0 R /R19 60 0 R /R27 79 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] /ColorSpace << /R31 18 0 R >> >> /MediaBox [ 0 0 612 792 ] /Annots [ 190 0 R 191 0 R ] >> endobj 6 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 192 0 R /Resources << /ExtGState << /R16 45 0 R >> /Font << /R108 154 0 R /R118 158 0 R /R112 163 0 R /R110 166 0 R /R104 178 0 R /R114 182 0 R /R106 185 0 R /F2 193 0 R /F1 194 0 R /R122 139 0 R /R17 53 0 R /R138 195 0 R /R19 60 0 R /R27 79 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 198 0 R 199 0 R 200 0 R 201 0 R 202 0 R 203 0 R 204 0 R ] >> endobj 7 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 205 0 R /Resources << /XObject << /R166 206 0 R /R174 207 0 R /R175 208 0 R /R171 209 0 R /R172 210 0 R /R173 211 0 R >> /ExtGState << /R29 41 0 R /R32 46 0 R /R16 45 0 R >> /Font << /R108 154 0 R /R118 158 0 R /R167 212 0 R /R164 216 0 R /R110 166 0 R /R120 142 0 R /R104 178 0 R /R114 182 0 R /R106 185 0 R /F2 219 0 R /F1 220 0 R /R17 53 0 R /R19 60 0 R /R23 64 0 R /R25 71 0 R /R169 221 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] /ColorSpace << /R31 18 0 R >> >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 ] >> endobj 8 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 244 0 R /Resources << /ExtGState << /R16 45 0 R >> /Font << /R202 245 0 R /R118 158 0 R /R206 249 0 R /R204 252 0 R /R110 166 0 R /R104 178 0 R /R114 182 0 R /F2 256 0 R /F1 257 0 R /R17 53 0 R /R19 60 0 R /R25 71 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 258 0 R 259 0 R 260 0 R 261 0 R 262 0 R 263 0 R 264 0 R 265 0 R 266 0 R 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 ] >> endobj 9 0 obj << /Parent 1 0 R /Rotate 0 /Contents 276 0 R /Resources << /XObject << /R264 277 0 R /R265 280 0 R /R266 281 0 R /R260 282 0 R /R261 283 0 R /R262 284 0 R /R263 285 0 R /R223 286 0 R /R224 287 0 R /R300 288 0 R /R240 289 0 R /R241 290 0 R /R229 291 0 R /R289 292 0 R /R284 293 0 R /R238 306 0 R /R237 296 0 R /R235 297 0 R /R234 295 0 R /R232 298 0 R /R231 299 0 R /R259 300 0 R /R258 301 0 R /R255 302 0 R /R257 303 0 R /R270 304 0 R /R299 345 0 R /R298 346 0 R /R256 305 0 R /R291 348 0 R /R290 349 0 R /R293 350 0 R /R292 351 0 R /R295 352 0 R /R294 353 0 R /R297 347 0 R /R296 354 0 R /R228 294 0 R >> /ExtGState << /R29 41 0 R /R221 278 0 R /R32 46 0 R /R16 45 0 R >> /Font << /F2 355 0 R /F1 356 0 R /R17 53 0 R /R19 60 0 R /R110 166 0 R /R227 307 0 R /R287 357 0 R /R285 361 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] /ColorSpace << /R31 18 0 R >> >> /Group 344 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 365 0 R 366 0 R 367 0 R 368 0 R 369 0 R 370 0 R 371 0 R 372 0 R ] /Type /Page >> endobj 10 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 373 0 R /Resources << /ExtGState << /R16 45 0 R >> /Font << /R202 245 0 R /F2 374 0 R /F1 375 0 R /R114 182 0 R /R17 53 0 R /R19 60 0 R /R110 166 0 R /R104 178 0 R /R27 79 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 376 0 R 377 0 R 378 0 R 379 0 R 380 0 R 381 0 R 382 0 R 383 0 R 384 0 R 385 0 R 386 0 R 387 0 R 388 0 R 389 0 R 390 0 R 391 0 R 392 0 R 393 0 R 394 0 R 395 0 R 396 0 R 397 0 R 398 0 R 399 0 R 400 0 R 401 0 R 402 0 R 403 0 R 404 0 R 405 0 R 406 0 R 407 0 R 408 0 R 409 0 R 410 0 R 411 0 R 412 0 R 413 0 R 414 0 R 415 0 R 416 0 R 417 0 R 418 0 R 419 0 R 420 0 R 421 0 R 422 0 R ] >> endobj 11 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 423 0 R /Resources << /ExtGState << /R16 45 0 R >> /Font << /F2 424 0 R /R19 60 0 R /F1 425 0 R /R27 79 0 R /R17 53 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ ] >> endobj 12 0 obj << /Type /Catalog /Pages 1 0 R >> endobj 13 0 obj << /Length 24289 >> stream q q q 0.10000 0 0 0.10000 0 0 cm /R16 gs 0 g q 10 0 0 10 0 0 cm BT /R17 14.34620 Tf 1 0 0 1 62.26480 675.06700 Tm [ (AMC\072) -250.00700 (Attention) -250.01100 (guided) -250.01300 (Multi\055modal) -249.98800 (Corr) 18.00990 (elation) -250.00700 (Lear) 14.98930 (ning) -249.99500 (f) 24.99230 (or) -249.99500 (Image) -250.00500 (Sear) 17.99460 (ch) ] TJ /R19 11.95520 Tf 32.65310 -37.85820 Td [ (Kan) -250.00800 (Chen) ] TJ /R21 7.97010 Tf 48.14300 4.33906 Td [ (1) -0.29866 ] TJ /R19 11.95520 Tf 28.64300 -4.33906 Td [ (T) 34.98880 (rung) -250.00500 (Bui) ] TJ /R21 7.97010 Tf 49.06330 4.33906 Td [ (2) -0.30019 ] TJ /R19 11.95520 Tf 28.64300 -4.33906 Td [ (Chen) -249.99500 (F) 15.01580 (ang) ] TJ /R21 7.97010 Tf 51.95700 4.33906 Td [ (2) -0.30019 ] TJ /R19 11.95520 Tf 28.64300 -4.33906 Td [ (Zhao) 24.99360 (wen) -250.00800 (W) 80 (ang) ] TJ /R21 7.97010 Tf 74.76680 4.33906 Td [ (2) -0.30019 ] TJ /R19 11.95520 Tf 28.64300 -4.33906 Td [ (Ram) -250.01000 (Ne) 24.98750 (v) 24.98340 (atia) ] TJ /R21 7.97010 Tf 62.15430 4.33906 Td [ (1) -0.30019 ] TJ -334.27200 -13.94800 Td [ (1) -0.29866 ] TJ /R19 11.95520 Tf 4.73320 -4.33789 Td [ (Uni) 24.99460 (v) 14.98620 (ersity) -249.98900 (of) -250.01400 (Southern) -249.99500 (California) ] TJ /R21 7.97010 Tf 184.47800 4.33789 Td [ (2) -0.30019 ] TJ /R19 11.95520 Tf 4.73203 -4.33789 Td [ (Adobe) -249.99700 (Research) ] TJ /R23 8.96640 Tf -231.70500 -13.94800 Td (kanchen\100usc\056edu\054) Tj /R25 8.96640 Tf 91.45700 0 Td [ (f) -0.90126 ] TJ /R23 8.96640 Tf 4.60820 0 Td [ (bui\054) -600.00200 (cfang\054) -600.01000 (zhawang) ] TJ /R25 8.96640 Tf 102.21600 0 Td [ (g) -0.90126 ] TJ /R23 8.96640 Tf 4.60781 0 Td [ (\100adobe\056com\054) -600.01300 (nevatia\100usc\056edu) ] TJ /R17 11.95520 Tf -180.43400 -41.04610 Td (Abstract) Tj /R27 9.96260 Tf -83.92810 -23.91910 Td [ (Given) -338.99600 (a) -338.01700 (user\047) 40.00310 (s) -338.99600 (query) 55.01780 (\054) -360.99600 (tr) 14.99140 (aditional) -338.00700 (ima) 10.01300 (g) 10.00320 (e) -339.00700 (sear) 36.98650 (c) 15.01220 (h) -338.99700 (systems) ] TJ -11.95470 -11.95590 Td [ (r) 14.98400 (ank) -334.01700 (ima) 10.01360 (g) 10.00320 (es) -333.98200 (accor) 36.98160 (ding) -334 (to) -334.01000 (its) -334.00200 (r) 37.01960 (ele) 15.01590 (vance) -333.98600 (to) -334.99100 (a) -334.01800 (single) -333.99600 (modal\055) ] TJ 11.95510 TL T* [ (ity) -392.98300 (\050e) 15.00980 (\056g) 14.98520 (\056\054) -428.98200 (ima) 10.01300 (g) 10.00320 (e) -394.01800 (content) -393.01200 (or) -392.98200 (surr) 44.99620 (ounding) -394.01500 (te) 20 (xt\051\056) -739.99200 (Nowadays\054) ] TJ T* [ (an) -318.01600 (incr) 36.98830 (easing) -318.98100 (number) -318.00600 (of) -318.99300 (ima) 10.01300 (g) 10.00320 (es) -317.98500 (on) -318.01600 (the) -318.99800 (Internet) -318.00600 (ar) 36.98650 (e) -319.01000 (avail\055) ] TJ T* [ (able) -320.01200 (with) -319.99300 (associated) -319.98700 (meta) -320.01400 (data) -320.00300 (in) -320.01300 (ric) 14.98650 (h) -319.98100 (modalities) -320.01500 (\050e) 15.00980 (\056g) 14.98520 (\056\054) -338.01700 (ti\055) ] TJ T* [ (tles\054) -258.01700 (k) 9.99404 (e) 29.99070 (ywor) 36.99440 (ds\054) -257.00200 (ta) 10.01060 (gs\054) -257.98300 (etc\056\051\054) -257.01600 (whic) 14.99870 (h) -255.99200 (can) -257.01000 (be) -255.99600 (e) 19.99180 (xploited) -256.00900 (for) -256.01900 (better) ] TJ 11.95590 TL T* [ (similarity) -279 (measur) 36.99510 (e) -279.01700 (with) -279.01800 (queries\056) -397.00900 (In) -279.01100 (this) -278.98800 (paper) 111.01800 (\054) -287.00800 (we) -279.00200 (le) 14.98030 (ver) 15.01470 (a) 10.00320 (g) 10.00320 (e) ] TJ 11.95510 TL T* [ (visual) -302.00200 (and) -302.01300 (te) 20 (xtual) -302.98900 (modalities) -302.01900 (for) -302.01000 (ima) 10.01300 (g) 10.00320 (e) -301.99400 (sear) 36.98650 (c) 15.01220 (h) -301.98400 (by) -303.00800 (learning) ] TJ T* [ (their) -314.00500 (corr) 36.98710 (elation) -315.01400 (with) -313.99300 (input) -313.99700 (query) 55.01780 (\056) -503.99200 (Accor) 36.98160 (ding) -314.00400 (to) -314.01400 (the) -314.99900 (intent) ] TJ T* [ (of) -326.99200 (query) 55.01780 (\054) -346.99800 (attention) -327 (mec) 15.01100 (hanism) -327.01300 (can) -326.99800 (be) -327.00300 (intr) 44.99740 (oduced) -328.01100 (t) 0.98758 (o) -328.01900 (adap\055) ] TJ T* [ (tively) -370.00800 (balance) -369.99800 (the) -370.00900 (importance) -369.98200 (of) -370.00400 (dif) 18.01660 (fer) 36.98400 (ent) -370.00900 (modalities\056) -671.01300 (W) 91.98710 (e) ] TJ T* [ (pr) 44.98390 (opose) -300.99200 (a) -301.00400 (no) 10.00810 (vel) -301.01000 (Attention) -301.00500 (guided) -300.98500 (Multi\055modal) -300.99900 (Corr) 37.00360 (elation) ] TJ 11.95630 TL T* [ (\050AMC\051) -271.98200 (learning) -271.01700 (method) -272.01200 (whic) 14.99870 (h) -271.98900 (consists) -271.99400 (of) -271.00100 (a) -271.98900 (jointly) -271.99900 (learned) ] TJ 11.95470 TL T* [ (hier) 14.99200 (ar) 36.98520 (c) 15.01280 (hy) -466 (of) -465.98800 (intr) 15.00360 (a) -465.99600 (and) -465.98600 (inter) 20.01750 (\055attention) -465.99800 (networks\056) -958 (Condi\055) ] TJ T* [ (tioned) -343.99600 (on) -342.99200 (query\047) 39.99700 (s) -344.01500 (intent\054) -367.00700 (intr) 15.00360 (a\055attention) -342.99400 (networks) -343.98700 (\050i\056e) 15.01960 (\056\054) -366.99300 (vi\055) ] TJ T* [ (sual) -397.00900 (intr) 15.00360 (a\055attention) -396.98500 (network) -397.00400 (and) -397.01700 (langua) 9.99098 (g) 10.00320 (e) -396.99700 (intr) 15.00240 (a\055attention) ] TJ T* [ (network\051) -254.99200 (attend) -254.99200 (on) -255.00600 (informative) -255.01100 (parts) -254.98700 (within) -255.01100 (eac) 15.01470 (h) -255.01100 (modality\073) ] TJ T* [ (a) -291.00500 (multi\055modal) -290.01800 (inter) 20.01750 (\055attention) -291.00800 (network) -290.00200 (pr) 44.98510 (omotes) -290.98800 (the) -291.00300 (impor) 19.98440 (\055) ] TJ 11.95590 TL T* [ (tance) -203.01600 (of) -203.01300 (the) -203.01700 (most) -203.00400 (query\055r) 36.98400 (ele) 15.01590 (vant) -203.01300 (modalities\056) -294.01700 (In) -202.98300 (e) 19.99180 (xperiments\054) ] TJ 11.95510 TL T* [ (we) -293 (e) 15.01280 (valuate) -292.98900 (AMC) -292.99200 (models) -292.98800 (on) -293 (the) -293.00300 (sear) 36.98650 (c) 15.01220 (h) -293.00500 (lo) 10.01060 (gs) -292.99800 (fr) 44.98510 (om) -293.00300 (two) -292.98500 (r) 37.01830 (eal) ] TJ T* [ (world) -210.98700 (ima) 10.01360 (g) 10.00320 (e) -210.00800 (sear) 36.98650 (c) 15.01220 (h) -211.01900 (engines) -211.00400 (and) -211.00900 (show) -210.01400 (a) -211.01900 <7369676e690263616e74> -211.01400 (boost) -210.99400 (on) ] TJ T* [ (the) -291.00200 (r) 14.98400 (anking) -292.00700 (of) -290.99800 (user) 20.00890 (\055clic) 20 (k) 9.99343 (ed) -291.00900 (ima) 10.01300 (g) 10.00320 (es) -291.98900 (in) -290.99800 (sear) 36.98650 (c) 15.01220 (h) -291.98600 (r) 37.01830 (esults\056) -434.01600 (Addi\055) ] TJ T* [ (tionally) 55.01040 (\054) -261.01300 (we) -259.00600 (e) 19.99180 (xtend) -259.00700 (AMC) -258.99700 (models) -258.99400 (to) -259.00300 (caption) -258.98600 (r) 14.98280 (anking) -258.99400 (task) -259 (on) ] TJ T* [ (COCO) -387 (dataset) -385.98300 (and) -387.01900 (ac) 15.01830 (hie) 14.98520 (ve) -387.00300 (competitive) -387.00900 (r) 37.01830 (esults) -387.00700 (compar) 36.99380 (ed) ] TJ 11.95590 TL T* [ (with) -250.00400 (r) 37.01960 (ecent) -249.99800 (state\055of\055the\055arts\056) ] TJ /R17 11.95520 Tf 25.97110 TL T* [ (1\056) -249.99000 (Intr) 18.01460 (oduction) ] TJ /R19 9.96260 Tf 11.95470 -18.93710 Td [ (Image) -445.99900 (search) -446.98900 (by) -445.99400 (te) 14.98030 (xt) -447.01100 (is) -445.99100 (widely) -445.98100 (used) -447.01100 (in) -445.99200 (e) 25.01050 (v) 14.98280 (eryday) -446.98200 (life) ] TJ -11.95470 -11.95590 Td (\050) Tj /R27 9.96260 Tf 3.31797 0 Td [ (e) 15.01280 (\056g) 14.98520 (\056) ] TJ /R19 9.96260 Tf 14.08670 0 Td [ (\054) -239.99700 (search) -236.98500 (engines\054) -240.01600 (security) -237 (surv) 14.99010 (eillance\054) -239.98700 (mobil) 1.01699 (e) -238.00400 (phones\051\056) ] TJ -17.40470 -11.95510 Td [ (Gi) 24.98600 (v) 14.98280 (en) -327.00400 (a) -327.00900 (te) 14.98100 (xtual) -326.98500 (query) 65.01610 (\054) -346.99800 (i) 0.98758 (mage) -328.01500 (search) -327.00900 (systems) -327.00400 (retrie) 25.01540 (v) 14.98280 (e) -327.00900 (a) -327.00900 (set) ] TJ 11.95470 TL T* [ (of) -315.00400 (related) -313.98400 (images) -314.98400 (by) -314.01600 (the) -314.99900 (rank) -315.00400 (of) -313.98600 (their) -314.99400 (rele) 25.01050 (v) 24.98110 (ance\056) -504 (Learning) ] TJ T* [ (this) -254.01000 (rele) 25.01050 (v) 24.98110 (ance\054) ] TJ /R27 9.96260 Tf 59.62500 0 Td [ (i\056e) 14.98280 (\056) ] TJ /R19 9.96260 Tf 12.02500 0 Td [ (\054) -255.01400 (correlation) -254.01900 (between) -253.98900 (query) -253.98900 (and) -253.99200 (image\054) -255 (is) ] TJ -71.65000 -11.95510 Td [ (k) 10.00320 (e) 15.01280 (y) -249.99300 (to) -249.98500 (the) -249.99000 (system\047) 55 (s) -249.99100 (utility) 65.01490 (\056) ] TJ 11.95470 -11.96410 Td [ (T) 79.99030 (o) -406.98600 (measure) -406.98400 (the) -406.98400 (correlation) -407.01300 (between) -406.98400 (q) -1.01454 (ue) 0.98023 (ry) -408.00800 (and) -406.98600 (image\054) ] TJ -11.95470 -11.95510 Td [ (typically) -236.00400 (a) -236.00400 (shared) -235.99600 (latent) -235.00100 (subspace) -235.98500 (is) -235.98600 (learned) -236 (for) -236.00200 (query\047) 55.01530 (s) -235.99500 (te) 14.98030 (xt) ] TJ 11.95590 TL T* [ (modality) -344.99600 (and) -346.01500 (a) -345.00600 (single) -345.01500 (image\055related) -344.98500 (modality) -346.01600 (\050) ] TJ /R27 9.96260 Tf 192.19400 0 Td [ (e) 15.01220 (\056g) 14.98520 (\056) ] TJ /R19 9.96260 Tf 14.08830 0 Td [ (\054) -368.99500 (visual) ] TJ -206.28200 -11.95510 Td [ (contents\054) -399.00200 (surrounding) -370.01400 (te) 14.98030 (xt\051\056) -667.99500 (T) 35.01740 (raditional) -369.98500 (image) -368.98300 (search) -369.98200 (en\055) ] TJ 11.95510 TL T* [ (gines) -449.00400 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 78.93400 92.95510 Tm (2) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 83.91520 92.95510 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 90.86910 92.95510 Tm (33) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 100.83200 92.95510 Tm [ (\135) -448.98800 (match) -449.00700 (queries) -447.99300 (with) -449.00100 (te) 14.98030 (xt) -447.99100 (or) -448.98100 (tags) -449.00800 (associated) ] TJ -50.71950 -11.95510 Td [ (with) -391.01100 (images\056) -732.01000 (DSSM) -390.99000 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 144.07000 81 Tm (11) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 154.03200 81 Tm [ (\135) -390.99700 (learns) -390.98700 (an) -390.99200 (embedding) -390.99600 (subspace) ] TJ ET Q /R29 gs q 3225.87000 4373.99000 2088.03000 1408.32000 re W n /R31 cs 1 1 1 scn 3225.87000 4373.99000 2088.03000 1408.32000 re f /R32 gs 0 0 0 scn q 10 0 0 10 0 0 cm BT -0.05271 Tc /R33 6.84600 Tf 1 0 0 1 434.57100 459.40000 Tm [ (\041\042) -22.99470 (\043) -22.97330 (\044) -15 (\045) -39.98400 (\046) 24 (\047) ] TJ 0.05271 Tc /R35 6.84600 Tf 29.34020 0 Td (\041\042) Tj ET Q 0 0.68945 0.31372 scn q 10 0 0 10 0 0 cm BT -0.16499 Tc /R33 6.84600 Tf 1 0 0 1 467.33400 459.40000 Tm [ (\050\046) 8.01677 (\042) -39.99830 (\051) -39.99830 (\052) -32.02430 (\047) -55.98190 (\042) -39.99830 (\053) -55.98190 (\054) ] TJ 0.05271 Tc -32.76290 -7.82383 Td [ (\055\056) -24.01470 (\057) -7.97398 (0) 40.01970 (\057) ] TJ ET Q 0 0 0 scn q 10 0 0 10 0 0 cm BT 0.05271 Tc /R35 6.84600 Tf 1 0 0 1 458.04300 451.57600 Tm (\043\042) Tj ET Q 0.75391 0 0 scn q 10 0 0 10 0 0 cm BT -0.05271 Tc /R33 6.84600 Tf 1 0 0 1 461.46600 451.57600 Tm [ (12) -39.99830 (\046) 24 (\052) -14.99220 (\051) -22.97330 (\054) -32.00290 (0) 23.99330 (\057) -22.97330 (\051) ] TJ 0.21839 Tc 35.20820 0 Td [ (2\0453) 24.01470 (\052) 23.97900 (\047\057) 16.98930 (\043) ] TJ ET Q 0 0 0 scn q 10 0 0 10 0 0 cm BT 0.05271 Tc /R35 6.84600 Tf 1 0 0 1 521.61300 451.57600 Tm (\043\042) Tj 0 Tc -87.04220 -8.31328 Td [ (\044) -7.69582 (\045\046) -15.44870 (\047) -24.12880 (\045\050) -24.07880 (\046) -15.38450 (\051) -15.38450 (\052) -24.07880 (\043) -7.69582 (\042) 63.68480 (\053) 7.68155 (\051) -15.38450 (\054) -15.38450 (\054) -15.38450 (\055\042) 63.67770 (\056) -24.07880 (\051) -15.38450 (\057) 7.68868 (0) 7.68868 (\051) -15.38450 (\042) 63.34250 (1) ] TJ 0.05271 Tc /R33 6.84600 Tf -107.79000 123.94100 Td [ (45) -24.01110 (\042) -7.97398 (\046) 40.02320 (\043) -7.97398 (6) ] TJ 0 Tc /R35 6.84600 Tf 23.96090 0 Td (\041) Tj ET Q Q /R32 gs q 3264.99000 4701.62000 963.32800 738.38700 re W n q 963.33000 0 0 875.30500 3264.99000 4564.70000 cm /R37 Do Q Q q 3225.87000 4373.99000 2088.03000 1408.32000 re W n /R31 cs 0 0 0 scn q 10 0 0 10 0 0 cm BT -0.05271 Tc /R33 6.84600 Tf 1 0 0 1 326.78100 459.40000 Tm [ (\041\042) -22.99470 (\043) -22.97330 (\044) -15.00290 (\045) -39.98040 (\046) 23.99680 (\047) ] TJ 0 Tc /R35 6.84600 Tf 29.34020 0 Td [ (\041) -7.69225 (\042) 63.68840 (2) 7.68155 (3) 24.11810 (\042) -7.69225 (\054) -15.38810 (\050) -24.07530 (\046) -15.38810 (40) 7.67798 (5) -15.38810 (\046) -15.38810 (6) -15.38810 (7) -7.69225 (\043) -7.69225 (\042) ] TJ ET Q 0.75391 0 0 scn q 10 0 0 10 0 0 cm BT -0.05271 Tc /R33 6.84600 Tf 1 0 0 1 326.78100 451.57600 Tm [ (12) -40.00180 (\046) 23.99680 (\052) -14.99580 (\051) -22.97330 (\054) -32.00290 (0) 23.98970 (\057) -22.97330 (\051) -22.97330 (7) 55.98550 (8) 32.02790 (\046) 23.99680 (\042) -22.97330 (\042) ] TJ ET Q 0 0 0 scn q 10 0 0 10 0 0 cm BT /R35 6.84600 Tf 1 0 0 1 376.17000 451.57600 Tm [ (\043) -7.69225 (\042) -7.69225 (\045\046) -15.39520 (\050) -24.07530 (\046) -15.38810 (\052) -24.08240 (8) -15.38810 (6) -15.38810 (\055) 71.37350 (\043) -7.69582 (\042) ] TJ -49.38910 -8.31328 Td [ (9) -7.69225 (\051) -15.38810 (\052) -24.08240 (0) 7.68512 (\057) 7.68512 (\055\042) 63.62420 (1) ] TJ 0.05271 Tc /R33 6.84600 Tf 0 108.98100 Td [ (45) -24.01110 (\042) -7.97398 (\046) 40.02320 (\043) -7.97398 (9) ] TJ 0 Tc /R35 6.84600 Tf 23.96090 0 Td (\041) Tj ET Q 9.78000 w 8 M /R31 CS 0 0.68945 0.31372 SCN 3817.56000 4955.90000 224.94100 244.50000 re S Q q 4345.68000 4701.62000 963.33200 738.38700 re W n q 1108.31000 0 0 738.39000 4225.68000 4701.62000 cm /R40 Do Q Q q 3225.87000 4373.99000 2088.03000 1408.32000 re W n 9.78000 w 8 M /R31 CS 0 0.68945 0.31372 SCN 4482.60000 5127.05000 224.93800 244.50000 re S 0.75391 0 0 SCN 3318.78000 5043.92000 278.73000 381.42200 re S 4.89000 w 0.65039 0.65039 0.65039 SCN 3565.73000 5638.05000 1618.59000 122.25000 re S /R31 cs 0 0.68945 0.31372 scn q 10 0 0 10 0 0 cm BT 0.10210 Tc /R41 8.80200 Tf 1 0 0 1 359.96200 567.09200 Tm [ (\041\042) -9.00341 (\043) 27.00190 (\042) -9.02006 (\044) -9.99085 (\045) 22.00930 (\046) 71.00100 (\047) 6.98139 (\050) 36.97890 (\042) -9.02006 (\051) 32.99030 (\042) ] TJ ET Q 0.35693 0.60742 0.83594 scn 5184.32000 5638.05000 127.13700 122.25000 re f 0.35693 0.60742 0.83594 SCN 5184.32000 5638.05000 127.13700 122.25000 re S 9.78000 w 1 1 1 SCN 5211.21000 5716.29000 m 5211.21000 5725.75000 5219.97000 5733.41000 5230.77000 5733.41000 c 5241.57000 5733.41000 5250.33000 5725.75000 5250.33000 5716.29000 c 5250.33000 5706.84000 5241.57000 5699.18000 5230.77000 5699.18000 c 5219.97000 5699.18000 5211.21000 5706.84000 5211.21000 5716.29000 c h S 5245.44000 5704.07000 m 5274.44000 5669.77000 l S 4.89000 w 0.65039 0.65039 0.65039 SCN 3565.73000 5491.36000 1618.59000 122.25000 re S 0.75391 0 0 scn q 10 0 0 10 0 0 cm BT 0.19892 Tc /R41 8.80200 Tf 1 0 0 1 359.96200 552.56200 Tm [ (\052\053) 47.99880 (\043) 38.00240 (\054) 28.99340 (\055) 25.01600 (\056) 24 (\051) 43.98800 (\042) 1.98042 (\055) ] TJ ET Q 0.35693 0.60742 0.83594 scn 5184.32000 5491.36000 127.13700 122.25000 re f 0.35693 0.60742 0.83594 SCN 5184.32000 5491.36000 127.13700 122.25000 re S 9.78000 w 1 1 1 SCN 5211.21000 5569.59000 m 5211.21000 5579.05000 5219.97000 5586.71000 5230.77000 5586.71000 c 5241.57000 5586.71000 5250.33000 5579.05000 5250.33000 5569.59000 c 5250.33000 5560.14000 5241.57000 5552.48000 5230.77000 5552.48000 c 5219.97000 5552.48000 5211.21000 5560.14000 5211.21000 5569.59000 c h S 5245.44000 5557.37000 m 5274.44000 5523.07000 l S Q q 10 0 0 10 0 0 cm BT /R19 8.96640 Tf 1 0 0 1 308.86200 426.44000 Tm [ (Figure) -527.01600 (1\056) -525.98700 (F) 15.00560 (or) -526.98100 (dif) 25.00110 (ferent) -526.98100 (queries\054) -595.99100 (it) -526.99500 (is) -525.99800 (helpful) -526.98400 (to) -527 (select) -525.99800 (query\055) ] TJ 10.95900 TL T* [ (dependent) -225.02100 (information) -224.98500 (within) -225 (and) -224.99400 (cross) -224.99400 (rich) -224.99400 (image\055related) -225.01500 (modal\055) ] TJ T* [ (ities) -370.99000 (a) 20.02100 (v) 25.00660 (ailable) -370.99000 (on) -371.00300 (the) -372.01600 (Internet) 0.99656 (\056) -673.97900 (Bounding) -370.98100 (box) 14.98650 (es) -371.01900 (and) -370.98100 (highlighted) ] TJ 10.95780 TL T* [ (k) 10.02000 (e) 15.01100 (yw) 10.00920 (ords) -249.98900 (correspond) -249.99700 (to) -250.01400 (dif) 25.00110 (ferent) -249.98900 (queries\047) -249.98900 (intent) -250.01700 (by) -249.97800 (their) -249.99500 (colors\056) ] TJ /R19 9.96260 Tf 24.39610 TL T* [ (to) -372.00400 (measure) -372.98900 (the) -372.00900 (correlation) -373.02100 (between) -372.01100 (document\055related) -373.00600 (te) 14.98280 (xt) ] TJ 11.95510 TL T* [ (modality) -231.99600 (and) -231.01500 (query\047) 55.01530 (s) -231.99600 (te) 14.98030 (xt) -231.01000 (modality) -231.99600 (using) -230.99100 (deep) -232 (learning\056) -304.00800 (On) ] TJ T* [ (the) -309 (other) -308.99700 (hand\054) -324.01200 (cross\055modal) -309.99700 (methods) -309.01200 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 464.69700 345.25800 Tm (39) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 474.66000 345.25800 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 480.22900 345.25800 Tm (7) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 485.21000 345.25800 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 490.77900 345.25800 Tm (32) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 500.74200 345.25800 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 506.31100 345.25800 Tm (5) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 511.29200 345.25800 Tm [ (\135) -309.99200 (l) 0.98513 (earn) -309.99200 (a) ] TJ -202.43000 -11.95630 Td [ (subspace) -231.98600 (to) -231.98800 (better) -231.99300 (measure) -231.99600 (correlation) -231.98600 (between) -231.99600 (query\047) 55.01530 (s) -231.99600 (te) 14.98280 (xt) ] TJ 11.95470 TL T* [ (modality) -233.99500 (and) -235.01500 (image) 0.99493 (\047) 53.98850 (s) -233.99500 (visual) -234.01500 (modality) 65.01370 (\056) -305.00300 (In) -234 (recent) -234.00500 (years\054) -238 (mul\055) ] TJ T* [ (tiple) -359.01600 (image\055related) -358.00400 (modalities) -358.98900 (are) -357.98900 (becoming) -359.00900 (widely) -359.01400 (a) 19.99180 (v) 24.98110 (ail\055) ] TJ T* [ (able) -372.01400 (online) -371.99200 (\050) ] TJ /R27 9.96260 Tf 52.23670 0 Td [ (e) 15.01220 (\056g) 14.98520 (\056) ] TJ /R19 9.96260 Tf 14.08710 0 Td [ (\054) -402.98900 (images) -371.99700 (on) -372.00700 (social) -372.00200 (netw) 10.00810 (orks) -372.00200 (are) -371.98700 (typically) ] TJ -66.32380 -11.95510 Td [ (posted) -369.99200 (with) -369.99500 (captions) -371.00900 (and) -370.01200 (tags\054) -399.99700 (follo) 24.99580 (wed) -370.01200 (by) -370.00700 (friends\047) -370.98700 (com\055) ] TJ T* [ (ments\051\056) -298.98400 (T) 69.99320 (e) 15.01220 (xt) -217.99100 (matching) -218.01000 (and) -219.01700 (cross\055m) 1 (odal) -219.00800 (methods) -218.00800 (are) -218.01300 (subop\055) ] TJ 11.95590 TL T* [ (timal) -214.99400 (due) -215.99800 (to) -215.01100 (their) -215.99100 (focus) -215.01600 (on) -215.99300 (onl) 1 (y) -215.99800 (single) -214.99800 (image\055related) -215.98800 (modal\055) ] TJ 11.95510 TL T* [ (ity) 64.98920 (\056) -400.98900 (As) -279.98300 (sho) 24.99340 (wn) -281.00500 (in) -280.01900 (Fig) ] TJ ET Q 1 0 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 394.26200 249.61600 Tm (1) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 399.24300 249.61600 Tm [ (\054) -287.98900 (image) -280.99700 (content) -280.01200 (can) -279.99700 (pro) 14.98280 (vide) -280.99700 (detailed) ] TJ -90.38090 -11.95470 Td [ (visual) -198.01800 (information) -198.00100 (\050) ] TJ /R27 9.96260 Tf 78.10900 0 Td [ (e) 15.01220 (\056g) 14.98520 (\056) ] TJ /R19 9.96260 Tf 14.08710 0 Td [ (\054) -208.98200 (col) 0.99003 (or) 39.01310 (\054) -208 (te) 14.98280 (xture\051) -198.00100 (of) -198.00600 (objects) -197.98700 (while) -197.98700 (k) 10.00320 (e) 15.01220 (y\055) ] TJ -92.19610 -11.95510 Td [ (w) 10 (ords) -334.99100 (can) -335.98600 (of) 25.01790 (fer) -335.01300 (abstract) -336.01300 (concepts) -334.98800 (\050) ] TJ /R27 9.96260 Tf 143.05700 0 Td [ (e) 15.01220 (\056g) 14.98770 (\056) ] TJ /R19 9.96260 Tf 14.08670 0 Td [ (\054) -356.99400 (scene) -335.00300 (description\051) ] TJ -157.14400 -11.95510 Td [ (or) -376.99300 (e) 15.01220 (xternal) -376.00300 (background) -377.01300 (information) -376.01100 (\050) ] TJ /R27 9.96260 Tf 152.63700 0 Td [ (e) 15.01220 (\056g) 14.98770 (\056) ] TJ /R19 9.96260 Tf 14.08790 0 Td [ (\054) -408.00500 (people\047) 54.98100 (s) -376.99100 (identi\055) ] TJ -166.72500 -11.95510 Td [ (ties\051\056) -627.99800 (Dif) 24.98360 (ferent) -355.98200 (modalities) -356.00900 (describe) -356.01400 (images) -356 (from) -356.00900 (dif) 24.98600 (ferent) ] TJ 11.95590 TL T* [ (vie) 24.98360 (ws\054) -302.01600 (which) -290.99600 (together) -290.99300 (pro) 14.98280 (vide) -292.01500 (inform) 1.00473 (ation) -292.01000 (in) -291 (a) -291.01500 (more) -291.98100 (com\055) ] TJ 11.95510 TL T* [ (prehensi) 24.99340 (v) 14.98280 (e) -305.01300 (w) 10 (ay) 65.00880 (\056) -475.99700 (It) -305.00300 <62656e65027473> -305.01800 (to) -306.01800 (learn) -305.00800 (a) -305.01300 (subspace) -304.99300 (to) -306.01800 (measure) ] TJ T* [ (the) -374.98900 (correlati) 0.98758 (on) -374.98600 (between) -374.98900 (query\047) 55.01530 (s) -374.01100 (te) 14.98280 (xt) -374.98600 (modality) -374.01100 (and) -374.99100 (image\055) ] TJ T* [ (related) -249.99500 (modalities\054) ] TJ /R27 9.96260 Tf 76.09300 0 Td [ (i\056e) 14.98280 (\056) ] TJ /R19 9.96260 Tf 12.02500 0 Td [ (\054) -249.99500 (multi\055modal) -250.01700 (correlation\056) ] TJ -76.16290 -13.19880 Td [ (There) -242.00100 (is) -241.00600 (a) -242.00400 (major) -241.98200 (challenge) -241.98900 (in) -241.00900 (learning) -241.97900 (this) -242.01400 (subspace\072) -306.00800 (not) ] TJ -11.95510 -11.95510 Td [ (all) -348.01100 (modalities) -346.99100 (are) -347.99300 (equally) -348 (informati) 25.01050 (v) 14.98280 (e) -347.98600 (due) -346.99600 (to) -348.01100 (the) -348.01100 (v) 24.98110 (ariation) ] TJ 11.95590 TL T* [ (in) -298.01600 (query\047) 55.01530 (s) -297.00400 (intent\056) -452.99300 (T) 79.99160 (o) -297.98500 (o) 14.98280 (v) 14.98280 (ercome) -297.99000 (this) -296.98500 (problem\054) -309.99700 (we) -297.99000 (introduce) ] TJ 11.95510 TL T* [ (an) -218.00300 (attention) -218.00800 (mechanism) -218.00800 (to) -217.99300 (adapti) 24.99090 (v) 14.98280 (ely) -217.99300 (e) 25.01050 (v) 24.98110 (aluate) -217.99800 (the) -217.99300 (rele) 25.01050 (v) 24.98110 (ance) ] TJ T* [ (between) -197.01900 (a) -198.01100 (modality) -196.98200 (and) -198.00100 (query\047) 55.01530 (s) -196.98200 (intent\056) -292 (F) 14.99260 (or) -198.00600 (the) -197.01600 (image) -197.99100 (search) ] TJ T* [ (task\054) -247 (we) -246 (consider) -245.98100 (tw) 10.00810 (o) -245.01300 (kinds) -246.00800 (of) -245.99800 (attention) -246.00300 (mechanisms\056) -309.00200 (First\054) ] TJ -13.74100 -29.88790 Td (1) Tj 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 47 72 14 re f* 0.50000 0.50000 0.50000 rg BT /F2 9 Tf 10.80000 TL ET BT 1 0 0 1 297 50 Tm (2644) Tj T* ET Q endstream endobj 14 0 obj << /Filter /FlateDecode /Resources << /ExtGState << /a0 << /CA 1 /ca 1 >> >> /XObject << /x18 15 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 15 0 obj << /Filter /FlateDecode /Resources 16 0 R /Length 107 /Type /XObject /BBox [ 78 746 96 765 ] /Subtype /Form >> stream xe AC̬wʠ =p,?]%+H-
Jc "82w8VSnGW;"
endstream
endobj
16 0 obj
<<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
>>
endobj
17 0 obj
<<
/Filter /DCTDecode
/Interpolate true
/Height 876
/Length 80397
/ColorSpace 18 0 R
/Width 962
/BitsPerComponent 8
/Subtype /Image
>>
stream
Adobe d C
$, !$4.763.22:ASF:=N>22HbINVX]^]8EfmeZlS[]Y C**Y;2;YYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYY l"
} !1AQa"q2#BR$3br
%&'()*456789:CDEFGHIJSTUVWXYZcdefghijstuvwxyz
w !1AQaq"2B #3Rbr
$4%&'()*56789:CDEFGHIJSTUVWXYZcdefghijstuvwxyz ? j08N1ޡgiF9qQK9ÎʘJzڀn
6fN0kմŸVg!]nqj2ٵ6ӱ!HޗnùIoj OՄ( RGחIi#ylC3b#F@=Q0pH@** W](`y#{$vwUOT]QaGZczǀxbHFs,}* oKrI`گ\r0jd