%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 12 0 R ] /Type /Pages /Count 10 >> endobj 2 0 obj << /Title (Generation and Comprehension of Unambiguous Object Descriptions) /Producer (PyPDF2) /Author (Junhua Mao\054 Jonathan Huang\054 Alexander Toshev\054 Oana Camburu\054 Alan L\056 Yuille\054 Kevin Murphy) /Subject (2016 IEEE Conference on Computer Vision and Pattern Recognition) >> endobj 3 0 obj << /Parent 1 0 R /Rotate 0 /Contents 14 0 R /Resources << /XObject << /x8 15 0 R /R55 18 0 R /x6 19 0 R /x12 22 0 R /R68 25 0 R /x10 51 0 R >> /ExtGState << /s9 54 0 R /R54 26 0 R /s5 57 0 R /R52 27 0 R /R53 28 0 R /s11 60 0 R /s7 63 0 R /a0 << /CA 1 /ca 1 >> /R33 66 0 R >> /Font << /F2 67 0 R /R57 36 0 R /R48 68 0 R /R46 72 0 R /R44 75 0 R /R50 80 0 R /R42 83 0 R /R40 87 0 R /R65 29 0 R /F1 91 0 R /R61 43 0 R /R38 92 0 R /R36 95 0 R /R34 99 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /Group 50 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 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 123 0 R 124 0 R ] /Type /Page >> endobj 4 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 125 0 R /Resources << /ExtGState << /R33 66 0 R >> /Font << /R36 95 0 R /F2 126 0 R /R34 99 0 R /F1 127 0 R /R48 68 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 128 0 R 129 0 R 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 142 0 R 143 0 R 144 0 R 145 0 R 146 0 R 147 0 R 148 0 R 149 0 R 150 0 R 151 0 R 152 0 R 153 0 R 154 0 R 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 166 0 R 167 0 R 168 0 R 169 0 R 170 0 R 171 0 R 172 0 R 173 0 R 174 0 R 175 0 R ] >> endobj 5 0 obj << /Parent 1 0 R /Rotate 0 /Contents 176 0 R /Resources << /XObject << /R145 177 0 R /R144 178 0 R /R146 179 0 R /R141 180 0 R /R140 181 0 R /R143 182 0 R /R142 183 0 R /R135 184 0 R /R136 185 0 R /R137 186 0 R /R138 187 0 R /R139 188 0 R >> /ExtGState << /R54 26 0 R /R33 66 0 R /R52 27 0 R /R53 28 0 R >> /Font << /F2 189 0 R /R147 190 0 R /R151 193 0 R /R44 75 0 R /R149 197 0 R /R65 29 0 R /F1 200 0 R /R61 43 0 R /R36 95 0 R /R34 99 0 R /R48 68 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /Group 50 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 201 0 R 202 0 R 203 0 R 204 0 R 205 0 R 206 0 R ] /Type /Page >> endobj 6 0 obj << /Parent 1 0 R /Rotate 0 /Contents 207 0 R /Resources << /XObject << /R202 208 0 R /R203 211 0 R /R200 212 0 R /R201 213 0 R /R206 214 0 R /R207 215 0 R /R204 216 0 R /R205 217 0 R /R220 218 0 R /R221 219 0 R /R222 220 0 R /R223 221 0 R /R224 222 0 R /R225 223 0 R /R226 224 0 R /R182 225 0 R /R180 226 0 R /R209 227 0 R /R181 228 0 R /R208 229 0 R /R211 230 0 R /R210 231 0 R /R213 232 0 R /R212 233 0 R /R215 234 0 R /R214 235 0 R /R217 236 0 R /R216 237 0 R /R219 238 0 R /R218 239 0 R /R235 240 0 R /R138 187 0 R /R178 255 0 R /R179 256 0 R /R183 257 0 R /R174 258 0 R /R175 259 0 R /R176 260 0 R /R177 261 0 R >> /ExtGState << /R54 26 0 R /R53 28 0 R /R33 66 0 R /R52 27 0 R /R173 209 0 R >> /Font << /R228 241 0 R /R192 262 0 R /R190 266 0 R /R44 75 0 R /R196 269 0 R /R184 274 0 R /R194 279 0 R /R186 282 0 R /R188 287 0 R /R198 291 0 R /R57 36 0 R /R42 83 0 R /R40 87 0 R /F2 296 0 R /F1 297 0 R /R48 68 0 R /R147 190 0 R /R151 193 0 R /R238 298 0 R /R149 197 0 R /R61 43 0 R /R232 248 0 R /R38 92 0 R /R36 95 0 R /R34 99 0 R /R236 301 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /Group 50 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 304 0 R 305 0 R 306 0 R 307 0 R 308 0 R 309 0 R 310 0 R 311 0 R 312 0 R 313 0 R 314 0 R 315 0 R 316 0 R 317 0 R ] /Type /Page >> endobj 7 0 obj << /Parent 1 0 R /Rotate 0 /Contents 318 0 R /Resources << /XObject << /R267 319 0 R /R268 320 0 R /R269 321 0 R /R138 187 0 R /R273 322 0 R /R272 323 0 R /R271 324 0 R /R270 325 0 R /R274 326 0 R >> /ExtGState << /R54 26 0 R /R53 28 0 R /R33 66 0 R /R52 27 0 R /R173 209 0 R >> /Font << /R264 327 0 R /R192 262 0 R /R190 266 0 R /R44 75 0 R /R196 269 0 R /R184 274 0 R /R194 279 0 R /R186 282 0 R /R188 287 0 R /R198 291 0 R /R57 36 0 R /R42 83 0 R /R40 87 0 R /F2 334 0 R /F1 335 0 R /R259 336 0 R /R48 68 0 R /R257 340 0 R /R261 344 0 R /R147 190 0 R /R151 193 0 R /R238 298 0 R /R149 197 0 R /R61 43 0 R /R38 92 0 R /R236 301 0 R /R36 95 0 R /R34 99 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /Group 50 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 348 0 R 349 0 R 350 0 R 351 0 R 352 0 R 353 0 R 354 0 R 355 0 R 356 0 R 357 0 R ] /Type /Page >> endobj 8 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 358 0 R /Resources << /ExtGState << /R33 66 0 R >> /Font << /F2 359 0 R /R190 266 0 R /F1 360 0 R /R261 344 0 R /R151 193 0 R /R36 95 0 R /R34 99 0 R /R44 75 0 R /R147 190 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 361 0 R 362 0 R 363 0 R 364 0 R 365 0 R 366 0 R 367 0 R 368 0 R ] >> endobj 9 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 369 0 R /Resources << /ExtGState << /R33 66 0 R >> /Font << /R190 266 0 R /R184 274 0 R /R238 298 0 R /R188 287 0 R /R44 75 0 R /R147 190 0 R /F2 370 0 R /F1 371 0 R /R38 92 0 R /R151 193 0 R /R36 95 0 R /R34 99 0 R /R261 344 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 372 0 R 373 0 R 374 0 R 375 0 R 376 0 R 377 0 R 378 0 R ] >> endobj 10 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 379 0 R /Resources << /XObject << /R361 380 0 R /R360 382 0 R /R363 383 0 R /R362 384 0 R /R365 385 0 R /R364 386 0 R /R367 387 0 R /R366 388 0 R /R369 389 0 R /R368 390 0 R /R329 391 0 R /R328 392 0 R /R325 393 0 R /R324 394 0 R /R327 395 0 R /R326 396 0 R /R321 397 0 R /R320 398 0 R /R323 399 0 R /R322 400 0 R /R315 401 0 R /R372 402 0 R /R373 403 0 R /R370 404 0 R /R371 405 0 R /R376 406 0 R /R377 407 0 R /R374 408 0 R /R375 409 0 R /R378 410 0 R /R379 411 0 R /R338 412 0 R /R339 413 0 R /R336 414 0 R /R337 415 0 R /R334 416 0 R /R335 417 0 R /R332 418 0 R /R333 381 0 R /R330 432 0 R /R331 433 0 R /R314 419 0 R /R349 434 0 R /R348 436 0 R /R347 437 0 R /R346 435 0 R /R345 438 0 R /R344 439 0 R /R343 440 0 R /R342 441 0 R /R341 442 0 R /R340 443 0 R /R309 428 0 R /R308 429 0 R /R307 430 0 R /R306 431 0 R /R383 444 0 R /R382 445 0 R /R381 446 0 R /R380 447 0 R /R387 448 0 R /R386 461 0 R /R385 449 0 R /R384 450 0 R /R354 451 0 R /R358 457 0 R /R359 453 0 R /R316 420 0 R /R317 421 0 R /R310 422 0 R /R311 423 0 R /R312 424 0 R /R313 425 0 R /R350 454 0 R /R351 455 0 R /R352 452 0 R /R353 456 0 R /R318 426 0 R /R319 427 0 R /R356 459 0 R /R355 458 0 R /R357 460 0 R >> /ExtGState << /R54 26 0 R /R53 28 0 R /R33 66 0 R /R52 27 0 R /R173 209 0 R >> /Font << /R48 68 0 R /F2 462 0 R /F1 463 0 R /R61 43 0 R /R36 95 0 R /R34 99 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /Group 50 0 R /MediaBox [ 0 0 612 792 ] /Annots [ ] >> endobj 11 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 464 0 R /Resources << /ExtGState << /R33 66 0 R >> /Font << /R48 68 0 R /R46 72 0 R /F2 465 0 R /F1 466 0 R /R36 95 0 R /R34 99 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 467 0 R 468 0 R 469 0 R 470 0 R 471 0 R 472 0 R 473 0 R 474 0 R 475 0 R 476 0 R 477 0 R 478 0 R 479 0 R 480 0 R 481 0 R 482 0 R 483 0 R 484 0 R 485 0 R 486 0 R 487 0 R 488 0 R 489 0 R 490 0 R 491 0 R 492 0 R 493 0 R 494 0 R 495 0 R 496 0 R 497 0 R 498 0 R 499 0 R 500 0 R 501 0 R 502 0 R 503 0 R 504 0 R 505 0 R 506 0 R 507 0 R 508 0 R 509 0 R 510 0 R 511 0 R 512 0 R 513 0 R 514 0 R 515 0 R 516 0 R 517 0 R ] >> endobj 12 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 518 0 R /Resources << /ExtGState << /R33 66 0 R >> /Font << /R36 95 0 R /F2 519 0 R /F1 520 0 R /R48 68 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 521 0 R 522 0 R 523 0 R 524 0 R 525 0 R 526 0 R 527 0 R 528 0 R 529 0 R 530 0 R 531 0 R 532 0 R 533 0 R 534 0 R 535 0 R 536 0 R 537 0 R 538 0 R 539 0 R 540 0 R 541 0 R 542 0 R 543 0 R 544 0 R 545 0 R 546 0 R 547 0 R 548 0 R 549 0 R 550 0 R ] >> endobj 13 0 obj << /Type /Catalog /Pages 1 0 R >> endobj 14 0 obj << /Length 22506 >> stream q q q 0.10000 0 0 0.10000 0 0 cm /R33 gs 0 g q 10 0 0 10 0 0 cm BT /R34 14.34620 Tf 1 0 0 1 84.53710 675.06700 Tm [ (Generation) -250.00900 (and) -249.99100 (Compr) 17.99120 (ehension) -249.98600 (of) -250.01200 (Unambiguous) -250 (Object) -249.99800 (Descriptions) ] TJ /R36 11.95520 Tf -28.44730 -20.85000 Td [ (Junhua) -249.99400 (Mao) ] TJ /R38 7.97010 Tf 58.77110 4.33906 Td [ (2) -0.29866 ] TJ /R40 7.97010 Tf 4.73203 0 Td [ <03> -0.29866 ] TJ /R36 11.95520 Tf 8.96602 -4.33906 Td [ (Jonathan) -249.99300 (Huang) ] TJ /R38 7.97010 Tf 77.36210 4.33906 Td [ (1) -0.30019 ] TJ /R36 11.95520 Tf 10.70980 -4.33906 Td [ (Ale) 15 (xander) -249.98700 (T) 80.01470 (oshe) 25.01400 (v) ] TJ /R38 7.97010 Tf 86.54340 4.33906 Td [ (1) -0.30019 ] TJ /R36 11.95520 Tf 10.70980 -4.33906 Td [ (Oana) -250.01600 (Camb) 19.99860 (uru) ] TJ /R38 7.97010 Tf 72.47110 4.33906 Td [ (3) -0.30019 ] TJ /R36 11.95520 Tf 10.70980 -4.33906 Td [ (Alan) -249.99700 (Y) 110.99400 (uille) ] TJ /R38 7.97010 Tf 54.79020 4.33906 Td [ (2) -0.30019 ] TJ /R42 7.97010 Tf 4.23398 0 Td [ (\073) -0.09802 ] TJ /R38 7.97010 Tf 2.35312 0 Td [ (4) -0.30019 ] TJ /R36 11.95520 Tf 10.70980 -4.33906 Td [ (K) 24.99570 (e) 24.98750 (vin) -250.00200 (Murph) 4.98688 (y) ] TJ /R38 7.97010 Tf 70.06800 4.33906 Td [ (1) -0.30019 ] TJ -476.44200 -14.32810 Td [ (1) -0.30019 ] TJ /R36 10.95890 Tf 4.73203 -3.95820 Td [ (Google) -250 (Inc\056) ] TJ /R38 7.97010 Tf 62.69490 3.95820 Td [ (2) -0.29866 ] TJ /R36 10.95890 Tf 4.73281 -3.95820 Td [ (Uni) 24.99350 (v) 14.98520 (ersity) -249.98500 (of) -249.98900 (California\054) -250.01500 (Los) -250.00400 (Angeles) ] TJ /R38 7.97010 Tf 176.40300 3.95820 Td [ (3) -0.30019 ] TJ /R36 10.95890 Tf 4.73320 -3.95820 Td [ (Uni) 24.99350 (v) 14.98410 (ersity) -249.98400 (of) -249.98900 (Oxford) ] TJ /R38 7.97010 Tf 103.04600 3.95820 Td [ (4) -0.30019 ] TJ /R36 10.95890 Tf 4.73203 -3.95820 Td [ (Johns) -249.98400 (Hopkins) -249.98900 (Uni) 24.99130 (v) 14.98410 (ersity) ] TJ /R44 9.96260 Tf -307.64200 -13.94800 Td (f) Tj /R46 9.96260 Tf 4.98203 0 Td (mjhustc\100\054yuille\100stat\056) Tj /R44 9.96260 Tf 125.52900 0 Td (g) Tj /R46 9.96260 Tf 4.98086 0 Td (ucla\056edu) Tj /R36 9.96260 Tf 47.82110 0 Td (\054) Tj /R46 9.96260 Tf 4.98086 0 Td (oana\055maria\056camburu\100cs\056ox\056ac\056uk) Tj /R44 9.96260 Tf -130.51000 -13.94770 Td (f) Tj /R46 9.96260 Tf 4.98086 0 Td (jonathanhuang\054toshev\054kpmurphy) Tj /R44 9.96260 Tf 173.35000 0 Td (g) Tj /R46 9.96260 Tf 4.98125 0 Td (\100google\056com) Tj /R34 11.95520 Tf -211.31100 -29.70820 Td (Abstract) Tj /R48 9.96260 Tf -83.92770 -15.97190 Td [ (W) 91.98650 (e) -363.00300 (pr) 44.98390 (opose) -362 (a) -362.99300 (method) -361.99700 (that) -363.01300 (can) -362.99200 (g) 10.00320 (ener) 15.01960 (ate) -362.01100 (an) -362.98800 (unambigu\055) ] TJ -11.95510 -11.95510 Td [ (ous) -268.99800 (description) -269.00800 (\050known) -268.99300 (as) -269.00300 (a) -269.00900 (r) 37.01960 (eferring) -268.98600 (e) 19.99180 (xpr) 36.98160 (ession\051) -268.99200 (of) -269.00200 (a) -269.00900 (spe\055) ] TJ 11.95590 TL T* [ <63690263> -249.99400 (object) -251.00100 (or) -249.98600 (r) 37.01960 (e) 39.98840 (gion) -250.99500 (in) -249.98500 (an) -249.98800 (ima) 10.01300 (g) 10.00320 (e) 9.99343 (\054) -251.01400 (and) -249.98300 (whic) 15 (h) -251.01200 (can) -249.99300 (also) -250.99500 (com\055) ] TJ 11.95510 TL T* [ (pr) 36.98520 (ehend) -301.00200 (or) -300.99800 (interpr) 37.00610 (et) -301.00600 (suc) 14.98520 (h) -301.00400 (an) -300.99900 (e) 19.99180 (xpr) 36.98160 (ession) -300.98400 (to) -302.01600 (infer) -300.98700 (whic) 15 (h) -301.00400 (ob\055) ] TJ T* [ (ject) -330.01800 (is) -330.01000 (being) -330.00600 (described\056) -548.98300 (W) 91.98590 (e) -329.98900 (show) -329.99400 (that) -329.99900 (our) -330.00600 (method) -330.00400 (outper) 19.98200 (\055) ] TJ T* [ (forms) -383.00700 (pr) 36.98520 (e) 15.01280 (vious) -383.01600 (methods) -383.00700 (that) -383.01000 (g) 10.00320 (ener) 15.01960 (ate) -382.98800 (descriptions) -382.98300 (of) -382.98300 (ob\055) ] TJ T* [ (jects) -277 (wit) 1.00780 (hout) -276.99100 (taking) -276.98800 (into) -276.00800 (account) -276.99400 (other) -276.01300 (potentially) -277.00500 (ambigu\055) ] TJ 11.95590 TL T* [ (ous) -342.00500 (objects) -342 (in) -342.98900 (the) -342.01400 (scene) 15.01470 (\056) -586.99800 (Our) -342.00800 (model) -342.00500 (is) -342.00900 (inspir) 37.01100 (ed) -341.98200 (by) -343.00200 (r) 37.01830 (ecent) ] TJ 11.95510 TL T* [ (successes) -340.99000 (of) -340.00900 (deep) -341 (learning) -341.00600 (methods) -339.99500 (for) -340.98500 (ima) 10.01300 (g) 10.00320 (e) -341.00700 (captioning) 10.00570 (\054) ] TJ T* [ (b) 20.00160 (ut) -275.98100 (while) -276.00400 (ima) 10.01300 (g) 10.00320 (e) -275.99800 (captioning) -275.98500 (is) -276.01900 (dif) 18.01660 <0263756c74> -276.01300 (to) -275.98100 (e) 15.01220 (valuate) 10.01790 (\054) -283.00900 (our) -276.01500 (task) ] TJ T* [ (allows) -241.01400 (for) -241 (easy) -241.01600 (objective) -241.00400 (e) 15.01220 (valuation\056) -307.00100 (W) 91.98710 (e) -240.98400 (also) -242.01600 (pr) 38.00590 (esent) -241.98900 (a) -241.01400 (ne) 15.01710 (w) ] TJ T* [ (lar) 36.99260 (g) 10.00320 (e\055scale) -253.01500 (dataset) -254.00500 (for) -252.99800 (r) 37.01960 (eferring) -252.98800 (e) 19.99180 (xpr) 36.98160 (essions\054) -254.00200 (based) -253 (on) -253.98700 (MS\055) ] TJ T* [ (COCO\056) -204.00900 (W) 91.98650 (e) -204.00900 (have) -204.00400 (r) 37.01960 (eleased) -203.99500 (the) -203.99800 (dataset) -204.01400 (and) -203.99000 (a) -204 (toolbox) -204.01500 (for) -203.98800 (visu\055) ] TJ 11.95590 TL T* [ (alization) -384 (and) -385.01900 (e) 15.01220 (valuation\054) -416.98300 (see) ] TJ ET Q 0 1 0 0 k q 10 0 0 10 0 0 cm BT /R50 9.96260 Tf 1 0 0 1 172.70800 411.27600 Tm [ (https\072\057\057) -3.98218 (github\056com\057) ] TJ -122.59600 -11.95510 Td [ (mjhucla\057) -3.98279 (Google\137Refexp\137toolbox) ] TJ ET Q 0 g q 10 0 0 10 0 0 cm BT /R48 9.96260 Tf 1 0 0 1 224.05000 399.32100 Tm (\056) Tj /R34 11.95520 Tf -173.93800 -23.39100 Td [ (1\056) -249.99000 (Intr) 18.01460 (oduction) ] TJ /R36 9.96260 Tf 11.95510 -13.82700 Td [ (There) -275.99600 (has) -275.00600 (been) -275.99200 (a) -275.99800 (lot) -274.99300 (of) -275.99200 (recent) -275.99700 (interest) -274.99300 (in) -275.98100 (generating) -276.01500 (te) 14.98030 (xt) ] TJ -11.95510 -11.95510 Td [ (descriptions) -214.00100 (of) -214.00200 (images) -215 (\050see) -214.01000 (e\056g\056\054) -221.01300 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 181.75800 350.14800 Tm (13) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 191.72100 350.14800 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 196.34300 350.14800 Tm (53) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 206.30600 350.14800 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 210.93800 350.14800 Tm (9) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 215.92000 350.14800 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 220.54200 350.14800 Tm (5) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 225.52300 350.14800 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 230.14600 350.14800 Tm (12) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 240.10900 350.14800 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 244.73100 350.14800 Tm (26) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 254.69400 350.14800 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 259.31600 350.14800 Tm (28) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 269.27900 350.14800 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 273.91100 350.14800 Tm (40) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 283.87400 350.14800 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 50.11210 338.19300 Tm (55) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 60.07460 338.19300 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 64.52770 338.19300 Tm (8) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 69.50900 338.19300 Tm [ (\135\051\056) -293.01400 (Ho) 24.98420 (we) 25.01420 (v) 14.98280 (er) 39.98600 (\054) -208.00200 (fundamentally) -196.98300 (this) -197.98000 (problem) -196.99900 (of) -196.98400 (image) -197.99100 (cap\055) ] TJ -19.39690 -11.95470 Td [ (tioning) -204.00300 (is) -203.99200 (subjecti) 25.00560 (v) 14.98280 (e) -204.01000 (and) -204 (ill\055posed\056) -295.01100 (W) 39.99460 (ith) -203.98600 (so) -203.99300 (man) 14.99010 (y) -204.98000 (v) 24.98110 (alid) -203.99000 (w) 10 (ays) ] TJ 11.95510 TL T* [ (to) -249.00500 (describe) -249.01100 (an) 15.01830 (y) -249.01200 (gi) 24.98850 (v) 14.98280 (en) -249.01600 (image\054) -250.01800 (automatic) -248.98500 (captioning) -249.02000 (methods) ] TJ T* [ (are) -268.00600 (thus) -268.01000 (notoriously) -268.01900 (dif) 24.98480 <0263756c74> -268.00400 (to) -269.00200 (e) 25.01050 (v) 24.98110 (aluate\056) -364.00800 (In) -267.99200 (particular) 40.00560 (\054) -273.01100 (ho) 24.98600 (w) ] TJ 11.95590 TL T* [ (can) -232.00400 (we) -232.00200 (decide) -231.99700 (that) -233.00500 (one) -231.99400 (sentence) -231.99400 (is) -231.98700 (a) -232.00500 (better) -231.99300 (description) -232.98300 (of) -231.99800 (an) ] TJ 11.95510 TL T* [ (image) -249.98200 (than) -249.98500 (another\077) ] TJ 11.95510 -12.52110 Td [ (In) -291.00800 (this) -292.00400 (paper) 39.99090 (\054) -301.00600 (we) -291.99200 (focus) -291.00200 (on) -291 (a) -291.99500 (speci) 0.99003 (al) -291.98800 (case) -291.01800 (of) -291.00800 (te) 14.98030 (xt) -292.01700 (genera\055) ] TJ -11.95510 -11.95590 Td [ (tion) -316.00100 (gi) 24.98850 (v) 14.98280 (en) -315.98600 (images\054) -331.99900 (where) -315.99000 (the) -316.01800 (goal) -314.99400 (is) -316.01200 (to) -316.01300 (generate) -315.98900 (an) -315.98600 (unam\055) ] TJ T* [ (biguous) -212.98500 (te) 14.98100 (xt) -211.99200 (description) -212.98800 (that) -211.98900 (applies) -213.00100 (to) -211.99200 (e) 15.01220 (xactly) -213.00900 (one) -211.99900 (object) -213.00900 (or) ] TJ T* [ (re) 15.00980 (gion) -201.00300 (in) -201.99300 (the) -201.01700 (image\056) -294.01100 (Such) -201.00600 (a) -200.99100 (description) -202.00900 (is) -201.01300 (kno) 24.99090 (wn) -201.01800 (as) -200.98400 (a) -202.01000 (\223refer) 20.01380 (\055) ] TJ T* [ (ring) -247.98400 (e) 15.01280 (xpression\224) -246.99100 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 121.29500 218.07500 Tm (50) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 131.25700 218.07500 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 136.21800 218.07500 Tm (52) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 146.18100 218.07500 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 151.13200 218.07500 Tm (41) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 161.09500 218.07500 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 166.05600 218.07500 Tm (42) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 176.01900 218.07500 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 180.98000 218.07500 Tm (14) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 190.94300 218.07500 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 195.89400 218.07500 Tm (19) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 205.85700 218.07500 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 210.81800 218.07500 Tm (27) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 220.78000 218.07500 Tm [ (\135\056) -309.00700 (This) -248.01500 (approach) ] TJ -170.66800 -11.95510 Td [ (has) -239.01200 (a) -238.98400 (major) -239.00200 (adv) 24.98240 (antage) -239.01500 (o) 14.98280 (v) 14.98280 (er) -238.98700 (generic) -239.01800 (image) -239.00400 (captioning\054) -241.01900 (since) ] TJ 11.95510 TL T* [ (there) -301.00900 (is) -300.01500 (a) -301.01300 <77656c6c2d6465026e6564> -299.99800 (performance) -301.00900 (metric\072) -411.00700 (a) -299.99400 (referring) -301.01300 (e) 15.01220 (x\055) ] TJ 11.95590 TL T* [ (pression) -363.99100 (i) 0.98758 (s) -364.01200 (considered) -364 (to) -362.98600 (be) -364.01600 (good) -363.01800 (if) -364.01400 (it) -362.98400 (uniquely) -363.98300 (describes) ] TJ 11.95510 TL T* [ (the) -303 (rele) 25.01050 (v) 24.98110 (ant) -303.98100 (object) -302.99200 (or) -303.00700 (re) 15.00980 (gion) -304.00600 (within) -303.01300 (its) -302.98900 (conte) 14.98520 (xt\054) -316.99100 (such) -302.99600 (that) -304.01300 (a) ] TJ T* [ (listener) -377.01400 (can) -377 (comprehend) -377.01400 (the) -376.98900 (description) -376.99800 (and) -376.99100 (then) -376.98400 (reco) 15.00980 (v) 14.98280 (er) ] TJ T* [ (the) -316.01800 (location) -315.99800 (of) -315.98500 (the) -314.99900 (original) -315.99600 (object\056) -508.01400 (In) -315.98500 (addition\054) -331.98400 (because) -315.98900 (of) ] TJ T* [ (the) -318.01700 (discriminati) 24.98970 (v) 14.98280 (e) -319.01000 (nature) -317.98100 (of) -317.98500 (the) -318.99800 (task\054) -334.98700 (referring) -319.01000 (e) 15.01220 (xpressions) ] TJ T* [ (tend) -344.98800 (to) -346.00800 (be) -345 (more) -346.01100 (detailed) -344.98300 (\050and) -344.99900 (therefore) -345.98600 (more) -344.99100 (useful\051) -346.00800 (than) ] TJ 11.95630 TL T* [ (image) -302.01200 (captions\056) -464.99100 (Finally) 65.01120 (\054) -313.98400 (it) -302.01300 (is) -302.01500 (easier) -301.99100 (to) -300.99600 (collect) -302.01300 (training) -301.99900 (data) ] TJ ET Q 3.98000 w 0 G 501.12100 998.78900 m 1446.11000 998.78900 l S q 10 0 0 10 0 0 cm BT /R36 7.97010 Tf 1 0 0 1 64.45780 90.46480 Tm [ (The) -289.98200 (major) -291.01600 (part) -289.98200 (of) -290.97700 (this) -290 (w) 10.01210 (ork) -290.00900 (w) 10.01210 (as) -290.97700 (done) -289.98500 (while) -289.97500 (J\056) -290.98900 (Mao) -290.02100 (and) -291.00200 (O\056) -289.99100 (Camb) 19.99050 (uru) ] TJ -14.34570 -9.46484 Td [ (were) -250.01100 (interns) -249.99700 (at) -249.99200 (Google) -250.01700 (Inc\056) ] TJ ET Q q 3147.69000 4665.06000 2244.38000 1261.24000 re W n /R68 Do Q q 10 0 0 10 0 0 cm BT /R36 8.96640 Tf 1 0 0 1 308.86200 454.08800 Tm [ (Figure) -203.97900 (1\056) -202.99300 (Illustration) -204.01400 (of) -203.98700 (our) -202.98500 (generation) -203.98400 (and) -203.99500 (comprehension) -203.02000 (system\056) ] TJ 10.95900 TL T* [ (On) -283.01200 (the) -283.01200 (left) -283.01500 (we) -282.98500 (see) -283.99200 (tha) 0.99111 (t) -283.99500 (the) -283.01200 (system) -282.99600 (is) -282.99000 (gi) 25.01740 (v) 14.98650 (en) -283.02300 (an) -283.02300 (image) -283.02300 (and) -283.02300 (a) -283.01700 (re) 14.99740 (gion) ] TJ T* [ (of) -298.00100 (interest\073) -322.00300 (it) -298.01200 (describes) -297.99800 (it) -299.01400 (as) -298.00400 (\223the) -298.01500 (man) -297.99800 (who) -297.99800 (is) -298.02000 (touching) -297.98700 (his) -298.02000 (head\224\054) ] TJ T* [ (which) -264.98400 (is) -264.99500 (unambiguous) -266.01000 (\050unlik) 9.98193 (e) -264.98100 (other) -264.98700 (possible) -265.00300 (e) 15.01370 (xpressions\054) -269.00600 (such) -264.98100 (as) ] TJ 10.95780 TL T* [ (\223the) -212.01900 (man) -210.99800 (wearing) -211.99800 (blue\224\054) -219.01200 (which) -211.00900 (w) 10.00920 (ould) -211.98400 (be) -212.01100 (unclear\051\056) -296.99600 (On) -210.99800 (the) -212 (right) -210.98200 (we) ] TJ 10.95900 TL T* [ (see) -290.00400 (that) -291.01700 (the) -289.98200 (system) -291.00900 (is) -290.00100 (gi) 25.01470 (v) 14.98650 (en) -290.99300 (an) -289.99300 (image\054) -300.99300 (an) -289.99300 (e) 15.01370 (xpression\054) -300.98800 (and) -289.99300 (a) -290.99000 (set) -290.02000 (of) ] TJ T* [ (candidate) -292.99100 (re) 14.99740 (gions) -293.00700 (\050bounding) -292.98300 (box) 14.98650 (es\051\054) -304.00500 (and) -293 (it) -293.00500 (selects) -293.00500 (the) -291.98600 (re) 14.99740 (gion) -293.01600 (that) ] TJ T* [ (corresponds) -249.99500 (to) -250.01400 (the) -249.98900 (e) 15.01100 (xpression\056) ] TJ /R36 9.96260 Tf 19.15630 TL T* [ (to) -249.00500 (\223co) 15.01470 (v) 14.98280 (er\224) -248.01000 (the) -249.01000 (space) -249.02000 (of) -249.01500 (reasonable) -247.98800 (referring) -248.98300 (e) 15.01220 (xpressions) -249.01700 (for) ] TJ 11.95590 TL T* [ (a) -250.00200 (gi) 24.98850 (v) 14.98280 (en) -249.99700 (object) -249.98300 (than) -249.98500 (it) -249.98300 (is) -249.98500 (for) -249.99700 (a) -250.00200 (whole) -249.98300 (image\056) ] TJ 11.95510 -12.70510 Td [ (W) 79.98660 (e) -327.98900 (consider) -326.98700 (tw) 10.00810 (o) -328.01900 (problems\072) -465.02000 (\0501\051) ] TJ /R48 9.96260 Tf 130.79000 0 Td [ (description) -328.01900 (g) 10.00320 (ener) 15.01710 (ation) ] TJ /R36 9.96260 Tf 91.01410 0 Td (\054) Tj -233.75900 -11.95470 Td [ (in) -310.99500 (which) -312.01200 (we) -311.00900 (must) -311.99700 (generate) -311.01200 (a) -311.01200 (te) 14.98280 (xt) -312.01700 (e) 15.01220 (xpression) -311.01700 (that) -312.01200 (uniquely) ] TJ 11.95510 TL T* [ (pinpoints) -334.01000 (a) -335.00800 (highlighted) -334.00800 (object\057re) 14.99260 (gion) -333.99800 (in) -334.99300 (the) -334.01300 (image) -334.00800 (and) -334.99800 (\0502\051) ] TJ /R48 9.96260 Tf T* [ (description) -446 (compr) 36.98890 (ehension) ] TJ /R36 9.96260 Tf 109.77700 0 Td [ (\054) -495.01100 (in) -445.99400 (which) -445.98900 (we) -446.00400 (must) -446.01400 (automati\055) ] TJ -109.77700 -11.95590 Td [ (cally) -318.01600 (select) -318.01800 (an) -317.00600 (object) -318.01100 (gi) 24.98850 (v) 14.98280 (en) -317.98600 (a) -317.99100 (te) 14.98280 (xt) -316.99600 (e) 15.01220 (xpression) -317.99600 (that) -318.01100 (refers) -318 (to) ] TJ T* [ (this) -377.01100 (object) -378 (\050see) -377 (Figure) ] TJ ET Q 1 0 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 404.69300 273.78300 Tm (1) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 409.67400 273.78300 Tm [ (\051\056) -692.00200 (Most) -377.00600 (prior) -378.00600 (w) 10.00320 (ork) -376.99100 (in) -376.98600 (the) -378.00600 (litera\055) ] TJ -100.81200 -11.95510 Td [ (ture) -217.99800 (has) -217.01500 (focused) -217.99300 (e) 15.01220 (xclusi) 25.00320 (v) 14.98280 (ely) -217.99300 (on) -217.01300 (description) -218.00800 (generation) -217.98300 (\050e\056g\056\054) ] TJ 11.95510 TL (\133) ' ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 312.18000 249.87300 Tm (31) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 322.14200 249.87300 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 327.26300 249.87300 Tm (27) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 337.22500 249.87300 Tm [ (\135\051\056) -353.00500 (Golland) ] TJ /R48 9.96260 Tf 47.37970 0 Td [ (et) -263.99300 (al) ] TJ /R36 9.96260 Tf 17.58710 0 Td [ (\056) -263.99300 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 410.63000 249.87300 Tm (19) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 420.59300 249.87300 Tm [ (\135) -265.02000 (consider) -264.02000 (generation) -264.01500 (and) -265.01000 (com\055) ] TJ -111.73100 -11.95470 Td [ (prehension\054) -250.00700 (b) 20.00160 (ut) -249.98500 (the) 14.98520 (y) -249.99300 (do) -249.98800 (not) -250.01700 (process) -249.98800 (real) -250.00200 (w) 10.00320 (orld) -249.98300 (images\056) ] TJ 11.95510 -12.70630 Td [ (In) -273.99100 (this) -274.00800 (paper) 39.99090 (\054) -279.99000 (we) -273.99600 (jointly) -273.98900 (model) -274.00800 (both) -274.00800 (tasks) -274.01300 (of) -273.99300 (description) ] TJ -11.95510 -11.95510 Td [ (generation) -334.00600 (and) -334.01800 (comprehension\054) -354.99500 (using) -335.01300 (state\055of\055the\055art) -334.01300 (deep) ] TJ 11.95470 TL T* [ (learning) -254.01900 (approaches) -253.01200 (to) -253.98500 (handle) -253.00200 (real) -254.00200 (images) -252.99700 (and) -253.99200 (te) 14.98280 (xt\056) -320.99100 (Specif\055) ] TJ T* [ (ically) 64.98670 (\054) -235.01700 (our) -229.99300 (model) -230.99600 (is) -231.00800 (based) -231.01300 (upon) -229.98100 (recently) -231.01000 (de) 25.01540 (v) 14.98280 (eloped) -231.00600 (methods) ] TJ T* [ (that) -372.00200 (combine) -371.99200 (con) 39.99820 (v) 20.00160 (olutional) -370.99200 (neural) -372.01100 (netw) 10.00810 (orks) -372.00200 (\050CNNs\051) -371.99700 (with) ] TJ 11.95590 TL T* [ (recurrent) -410.01500 (neural) -408.98600 (netw) 10.00810 (orks) -409.99500 (\050RNNs\051\056) -788.98600 (W) 79.98660 (e) -410.01500 (demonstrate) -410.02000 (that) ] TJ 11.95510 TL T* [ (our) -221.99500 (model) -222.01700 (outperforms) -222.01000 (a) -222.00700 (baseline) -221.98300 (which) -221.98700 (generates) -221.99700 (referring) ] TJ T* [ (e) 15.01220 (xpressions) -351.00300 (without) -351 (re) 15.00980 (g) 4.98446 (ard) -351.00500 (to) -350.01000 (the) -350.99000 (listener) -351.02000 (who) -350.98600 (must) -351.01000 (com\055) ] TJ T* [ (prehend) -316.01800 (the) -314.99900 (e) 15.01220 (xpression\056) -507.99900 (W) 79.98660 (e) -315.01100 (also) -316.01100 (sho) 24.99090 (w) -315.98200 (that) -314.99200 (our) -316.02100 (model) -316.00100 (can) ] TJ T* [ (be) -372.99700 (trained) -374.00400 (in) -372.98400 (a) -373.00100 (semi\055supervised) -372.99200 (f) 9.99343 (ashion\054) -403.98700 (by) -374.00600 (automatically) ] TJ T* [ (generating) -250.02000 (descriptions) -249.99500 (for) -249.99700 (image) -249.98300 (re) 15.00730 (gions\056) ] TJ 11.95510 -12.70630 Td [ (Being) -319.99800 (able) -321.00300 (t) 0.98758 (o) -321 (generate) -319.99100 (and) -320.02000 (comprehend) -320.98600 (obj) 1 (ect) -321.00500 (descrip\055) ] TJ -11.95510 -11.95510 Td [ (tions) -399.00200 (is) -400 (critical) -399.01900 (in) -399.01900 (a) -400.01700 (number) -399.01200 (of) -400.01200 (applications) -398.98700 (that) -399.01700 (use) -400.00700 (nat\055) ] 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 1 301.50000 50 Tm (11) Tj T* ET Q endstream endobj 15 0 obj << /Filter /FlateDecode /Resources << /ExtGState << /a0 << /CA 1 /ca 1 >> >> /XObject << /x18 16 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 16 0 obj << /Filter /FlateDecode /Resources 17 0 R /Length 107 /Type /XObject /BBox [ 78 746 96 765 ] /Subtype /Form >> stream xe AC̬wʠ =p,?]%+H-
Jc "82w8VSnGW;"
endstream
endobj
17 0 obj
<<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
>>
endobj
18 0 obj
<<
/Filter /DCTDecode
/BitsPerComponent 8
/Height 480
/Length 26662
/ColorSpace /DeviceRGB
/Width 640
/Subtype /Image
>>
stream
Adobe d C
$, !$4.763.22:ASF:=N>22HbINVX]^]8EfmeZlS[]Y C**Y;2;YYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYY "
} !1AQa"q2#BR$3br
%&'()*456789:CDEFGHIJSTUVWXYZcdefghijstuvwxyz
w !1AQaq"2B #3Rbr
$4%&'()*56789:CDEFGHIJSTUVWXYZcdefghijstuvwxyz ? ((((ެi&F<!n8M 7LHQ{PDO_7;tRR=~9}tcQvr1ÐJʿJB>j͛VH ծAe;sO.t78"] (3oz9BJ84`CP2Vms&|9zN,.]u}8ZfNb3vM$!&Ť 9 h/j cz1H#)&`')-b1ۨ?|1mwgޣfy~sL4h!HQPx@c4S;~ҏҏZ FZSB{ @4&>}jLӻ1қ(=SOZ@s@+ց?gCA@Ê(gIZ N>_ZU