%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 (DenseCap\072 Fully Convolutional Localization Networks for Dense Captioning) /Producer (PyPDF2) /Author (Justin Johnson\054 Andrej Karpathy\054 Li Fei\055Fei) /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 /R64 22 0 R /R63 26 0 R /R62 29 0 R /x12 45 0 R /x10 48 0 R >> /ExtGState << /s9 51 0 R /R49 24 0 R /R48 25 0 R /R54 23 0 R /s5 54 0 R /s7 57 0 R /R61 27 0 R /a0 << /CA 1 /ca 1 >> /s11 60 0 R /R33 63 0 R >> /Font << /R44 64 0 R /R46 68 0 R /R57 30 0 R /R51 37 0 R /R42 72 0 R /F2 75 0 R /F1 76 0 R /R38 77 0 R /R36 81 0 R /R34 85 0 R /R40 88 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /Group 44 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 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 ] /Type /Page >> endobj 4 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 115 0 R /Resources << /ExtGState << /R33 63 0 R >> /Font << /F2 116 0 R /F1 117 0 R /R44 64 0 R /R127 118 0 R /R131 121 0 R /R133 125 0 R /R135 130 0 R /R137 134 0 R /R36 81 0 R /R139 138 0 R /R34 85 0 R /R129 141 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 176 0 R 177 0 R 178 0 R 179 0 R 180 0 R 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 ] >> endobj 5 0 obj << /Parent 1 0 R /Rotate 0 /Contents 199 0 R /Resources << /XObject << /R163 200 0 R /R162 201 0 R /R161 202 0 R /R160 203 0 R /R152 204 0 R /R156 205 0 R /R157 206 0 R /R164 207 0 R /R154 208 0 R /R158 212 0 R /R159 213 0 R /R155 209 0 R /R151 211 0 R /R153 210 0 R >> /ExtGState << /R54 23 0 R /R61 27 0 R /R33 63 0 R /R49 24 0 R /R48 25 0 R >> /Font << /F2 214 0 R /R44 64 0 R /F1 215 0 R /R57 30 0 R /R127 118 0 R /R131 121 0 R /R133 125 0 R /R137 134 0 R /R36 81 0 R /R34 85 0 R /R129 141 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /Group 44 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 216 0 R 217 0 R 218 0 R 219 0 R 220 0 R 221 0 R ] /Type /Page >> endobj 6 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 222 0 R /Resources << /ExtGState << /R33 63 0 R >> /Font << /F2 223 0 R /R191 224 0 R /F1 229 0 R /R42 72 0 R /R127 118 0 R /R131 121 0 R /R133 125 0 R /R135 130 0 R /R137 134 0 R /R36 81 0 R /R139 138 0 R /R34 85 0 R /R129 141 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 ] >> endobj 7 0 obj << /Parent 1 0 R /Rotate 0 /Contents 251 0 R /Resources << /XObject << /R228 252 0 R /R229 253 0 R /R206 254 0 R /R207 255 0 R /R204 256 0 R /R205 257 0 R /R220 258 0 R /R221 259 0 R /R222 260 0 R /R223 261 0 R /R224 262 0 R /R225 263 0 R /R226 264 0 R /R227 265 0 R /R248 266 0 R /R249 267 0 R /R246 270 0 R /R247 271 0 R /R244 272 0 R /R245 273 0 R /R242 274 0 R /R243 275 0 R /R240 276 0 R /R241 277 0 R /R217 278 0 R /R209 269 0 R /R216 279 0 R /R219 280 0 R /R218 281 0 R /R208 268 0 R /R211 282 0 R /R210 283 0 R /R213 284 0 R /R212 285 0 R /R215 286 0 R /R214 287 0 R /R239 288 0 R /R238 289 0 R /R237 290 0 R /R236 291 0 R /R235 292 0 R /R234 293 0 R /R233 297 0 R /R232 294 0 R /R231 295 0 R /R230 296 0 R /R258 298 0 R /R251 299 0 R /R250 300 0 R /R253 301 0 R /R252 302 0 R >> /ExtGState << /R54 23 0 R /R61 27 0 R /R33 63 0 R /R49 24 0 R /R48 25 0 R >> /Font << /F2 310 0 R /R36 81 0 R /R44 64 0 R /R255 303 0 R /R34 85 0 R /R57 30 0 R /R42 72 0 R /F1 311 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /Group 44 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 312 0 R 313 0 R 314 0 R 315 0 R 316 0 R 317 0 R ] /Type /Page >> endobj 8 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 318 0 R /Resources << /ExtGState << /R33 63 0 R >> /Font << /F2 319 0 R /F1 320 0 R /R280 321 0 R /R44 64 0 R /R40 88 0 R /R127 118 0 R /R131 121 0 R /R278 324 0 R /R36 81 0 R /R34 85 0 R /R129 141 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 327 0 R 328 0 R 329 0 R 330 0 R 331 0 R 332 0 R 333 0 R 334 0 R 335 0 R 336 0 R 337 0 R 338 0 R 339 0 R ] >> endobj 9 0 obj << /Parent 1 0 R /Rotate 0 /Contents 340 0 R /Resources << /XObject << /R299 341 0 R /R298 342 0 R /R295 344 0 R /R297 343 0 R /R296 345 0 R /R302 346 0 R /R301 347 0 R /R300 348 0 R >> /ExtGState << /R33 63 0 R >> /Font << /R127 118 0 R /R131 121 0 R /F1 349 0 R /R36 81 0 R /R34 85 0 R /R292 350 0 R /F2 353 0 R /R129 141 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /Group 354 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 355 0 R 356 0 R 357 0 R 358 0 R ] /Type /Page >> endobj 10 0 obj << /Parent 1 0 R /Rotate 0 /Contents 359 0 R /Resources << /XObject << /R314 360 0 R /R315 361 0 R /R316 362 0 R /R317 363 0 R /R312 364 0 R /R313 365 0 R /R318 366 0 R /R319 367 0 R /R321 368 0 R /R320 369 0 R /R323 370 0 R /R322 371 0 R >> /ExtGState << /R33 63 0 R >> /Font << /R36 81 0 R /F2 372 0 R /R34 85 0 R /R44 64 0 R /F1 373 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /Group 354 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 374 0 R 375 0 R 376 0 R ] /Type /Page >> endobj 11 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 377 0 R /Resources << /ExtGState << /R33 63 0 R >> /Font << /R36 81 0 R /F2 378 0 R /R34 85 0 R /R44 64 0 R /F1 379 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 423 0 R 424 0 R 425 0 R 426 0 R 427 0 R 428 0 R 429 0 R 430 0 R 431 0 R 432 0 R 433 0 R 434 0 R 435 0 R 436 0 R 437 0 R ] >> endobj 12 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 438 0 R /Resources << /ExtGState << /R33 63 0 R >> /Font << /R36 81 0 R /F2 439 0 R /F1 440 0 R /R44 64 0 R /R42 72 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 441 0 R 442 0 R 443 0 R 444 0 R 445 0 R 446 0 R 447 0 R 448 0 R 449 0 R 450 0 R 451 0 R 452 0 R 453 0 R 454 0 R 455 0 R 456 0 R 457 0 R 458 0 R 459 0 R 460 0 R 461 0 R 462 0 R 463 0 R 464 0 R 465 0 R 466 0 R 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 ] >> endobj 13 0 obj << /Type /Catalog /Pages 1 0 R >> endobj 14 0 obj << /Length 20482 >> 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 63.26910 675.06700 Tm [ (DenseCap\072) -310.00200 (Fully) -249.99800 (Con) 40.00710 (v) 9.99625 (olutional) -249.99600 (Localization) -250.00500 (Netw) 9.99455 (orks) -250.01300 (f) 24.99230 (or) -249.99500 (Dense) -249.99500 (Captioning) ] TJ /R36 11.95520 Tf 100.14200 -37.85820 Td [ (Justin) -249.99500 (Johnson) ] TJ /R38 7.97010 Tf 70.08010 4.33906 Td [ <03> -0.30019 ] TJ /R36 11.95520 Tf 34.77700 -4.33906 Td [ (Andrej) -249.99100 (Karpath) 4.98892 (y) ] TJ /R38 7.97010 Tf 80.61290 4.33906 Td [ <03> -0.30019 ] TJ /R36 11.95520 Tf 34.77730 -4.33906 Td [ (Li) -249.98500 (Fei\055Fei) ] TJ -216.29600 -13.94690 Td [ (Department) -250 (of) -250.01400 (Computer) -250.01400 (Science\054) -249.99300 (Stanford) -250 (Uni) 24.99570 (v) 14.98510 (ersity) ] TJ /R40 8.96640 Tf 15.35590 -13.94800 Td [ (f) -0.89990 ] TJ /R42 8.96640 Tf 4.60820 0 Td (jcjohns\054karpathy\054feifeili) Tj /R40 8.96640 Tf 134.49500 0 Td [ (g) -0.90126 ] TJ /R42 8.96640 Tf 4.60781 0 Td (\100cs\056stanford\056edu) Tj /R34 11.95520 Tf -180.43400 -41.04610 Td (Abstract) Tj /R44 9.96260 Tf -95.88280 -23.91090 Td [ (W) 91.98650 (e) -367.98200 (intr) 44.99860 (oduce) -366.99100 (the) -368.01000 (dense) -366.99000 (captioning) -368.00800 (task\054) -397.01500 (whic) 15 (h) -366.99300 (r) 37.01830 (equir) 36.99380 (es) -368.01500 (a) ] TJ 11.95510 TL T* [ (computer) -207.00200 (vision) -208.01800 (system) -207.01100 (to) -207.99200 (both) -207.00200 (localize) -208.01400 (and) -207.01000 (describe) -207.98700 (salient) ] TJ T* [ (r) 37.01960 (e) 39.98900 (gions) -288.00200 (in) -288.99800 (ima) 10.01300 (g) 10.00320 (es) -287.99000 (in) -288.99800 (natur) 15.00610 (al) -288.01800 (langua) 9.99098 (g) 10.00320 (e) 15.01220 (\056) -426.00500 (The) -288.01600 (dense) -289.00300 (caption\055) ] TJ T* [ (ing) -280.01500 (task) -280.99800 (g) 10.00320 (ener) 15.01960 (alizes) -280.00500 (object) -280.99600 (detection) -280.00800 (when) -280.01200 (the) -281.00500 (descriptions) ] TJ T* [ (consist) -327.01100 (of) -326.99200 (a) -328.01900 (single) -327.01800 (wor) 36.99870 (d\054) -346.99400 (and) -326.99000 (Ima) 10.00320 (g) 10.00320 (e) -327.00900 (Captioning) -327.01900 (when) -328.00400 (one) ] TJ 11.95630 TL T* [ (pr) 36.98520 (edicted) -217.99100 (r) 37.01960 (e) 39.98900 (gion) -217.98100 (co) 10 (ver) 9.99588 (s) -217.01600 (the) -217.99500 (full) -218.01500 (ima) 10.01300 (g) 10.00320 (e) 15.01220 (\056) -299.00600 (T) 92 (o) -217.99800 (addr) 36.99630 (ess) -217.99500 (the) -217.99500 (local\055) ] TJ 11.95510 TL T* [ (ization) -214 (and) -213.98800 (description) -213.99700 (task) -213.01000 (jointly) -214.00800 (we) -213.99500 (pr) 44.98510 (opose) -213.98600 (a) -213.99900 (Fully) -213.98400 (Con\055) ] TJ T* [ (volutional) -348.99000 (Localization) -349.00900 (Network) -349.00400 (\050FCLN\051) -349.98100 (ar) 36.98650 (c) 15.01220 (hitectur) 37.00360 (e) -349.00500 (that) ] TJ T* [ (pr) 44.98390 (ocesses) -233.01000 (an) -233.01000 (ima) 10.01300 (g) 10.00320 (e) -232.98600 (with) -232.98700 (a) -233.01500 (single) 9.98485 (\054) -235.99700 (ef) 18 <026369656e74> -233.00500 (forwar) 37.01590 (d) -233.01500 (pass\054) -236.99500 (r) 37.01830 (e\055) ] TJ T* [ (quir) 36.99750 (es) -390.01200 (no) -390.98400 (e) 19.99240 (xternal) -389.99100 (r) 37.01960 (e) 39.98840 (gions) -391.00500 (pr) 44.98390 (oposals\054) -424.98500 (and) -391.01800 (can) -390.00800 (be) -390.99400 (tr) 14.99010 (ained) ] TJ T* [ (end\055to\055end) -294.01200 (with) -295.01600 (a) -293.98500 (single) -294.98400 (r) 45.01820 (ound) -294.01000 (of) -294.99700 (optimization\056) -442.99000 (The) -294.99500 (ar) 36.98650 (c) 15.01220 (hi\055) ] TJ 11.95590 TL T* [ (tectur) 36.99080 (e) -377 (is) -378.00200 (composed) -377.01400 (of) -378.00300 (a) -376.99100 (Con) 40.01660 (volutional) -378.00600 (Network\054) -408.99100 (a) -378.01000 (no) 10.00810 (vel) ] TJ 11.95510 TL T* [ (dense) -475.01100 (localization) -474.99500 (layer) 111.01100 (\054) -531.00700 (and) -475.00400 (Recurr) 36.98650 (ent) -475.01100 (Neur) 14.99750 (al) -475.00700 (Network) ] TJ T* [ (langua) 9.99159 (g) 10.00320 (e) -430.99100 (model) -431.01000 (that) -431.00200 (g) 10.00320 (ener) 15.01960 (ates) -429.99300 (the) -431.01900 (label) -431.00700 (sequences\056) -852.98300 (W) 91.98710 (e) ] TJ T* [ (e) 15.01280 (valuate) -272.01200 (our) -272.01700 (network) -272.00500 (on) -271.98400 (the) -271.98600 (V) 73.99130 (isual) -270.98300 (Genome) -271.98600 (dataset\054) -277.99800 (whic) 15 (h) ] TJ T* [ (comprises) -387.98400 (94\054000) -387.98700 (ima) 10.01300 (g) 10.00320 (es) -388.01200 (and) -389.01800 (4\054100\054000) -388.01400 (r) 37.01830 (e) 39.98840 (gion\055gr) 45.00230 (ounded) ] TJ T* [ (captions\056) -713.01900 (W) 91.98650 (e) -385 (observe) -384.00600 (both) -383.99200 (speed) -384.98700 (and) -384 (accur) 14.98520 (acy) -384.99700 (impr) 44.99490 (o) 10.00320 (ve\055) ] TJ 11.95590 TL T* [ (ments) -326.00700 (o) 10.00320 (ver) -327.00700 (baselines) -325.99100 (based) -326.00700 (on) -326.01400 (curr) 36.98650 (ent) -326.99700 (state) -326.00200 (of) -326.01200 (the) -326.01700 (art) -326.98700 (ap\055) ] TJ 11.95510 TL T* [ (pr) 44.98390 (oac) 14.98340 (hes) -249.99100 (in) -249.98500 (both) -250.01500 (g) 10.00320 (ener) 15.01960 (ation) -250.00700 (and) -249.98300 (r) 37.01960 (etrie) 14.99010 (val) -249.99000 (settings\056) ] TJ /R34 11.95520 Tf 24.43480 TL T* [ (1\056) -249.99000 (Intr) 18.01460 (oduction) ] TJ /R36 9.96260 Tf 15.32930 TL T* [ (Our) -261.01100 (ability) -262 (to) -261.00300 (ef) 25.00810 (fortlessly) -262.00200 (point) -260.98600 (out) -260.99800 (and) -261.99100 (describe) -261.00800 (all) -261.98600 (aspects) ] TJ 11.95510 TL T* [ (of) -287.98900 (an) -289.01000 (image) -288.01400 (relies) -289.00200 (on) -287.98100 (a) -289.01600 (strong) -288.00600 (semantic) -288.98600 (understanding) -287.98600 (of) -289.00800 (a) ] TJ T* [ (visual) -319.99900 (scene) -321.00700 (and) -320.01900 (all) -320.01500 (of) -319.98500 (its) -320.98400 (elements\056) -520.98400 (Ho) 24.98360 (we) 25.01300 (v) 14.98280 (er) 39.98600 (\054) -337 (despite) -320.98300 (nu\055) ] TJ T* [ (merous) -363.99700 (potential) -363.98200 (applications\054) -392.98800 (this) -363.99200 (ability) -363.98300 (remains) -363.99800 (a) -363.98300 (chal\055) ] TJ 11.95590 TL T* [ (lenge) -416 (for) -416.01000 (our) -416.00200 (state) -415.99700 (of) -416.00800 (the) -416.98200 (art) -416.00900 (visual) -415.98300 (recognition) -416.01400 (systems\056) ] TJ 11.95510 TL T* [ (In) -434.98400 (the) -433.99800 (last) -435.01000 (fe) 25.00810 (w) -434.00400 (years) -434.98700 (there) -434.00600 (has) -435.01800 (been) -434.00500 <7369676e690263616e74> -435.01600 (progress) ] TJ T* [ (in) -362.98600 (image) -362.98200 <636c6173736902636174696f6e> -364.00900 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 148.41300 217.66900 Tm (39) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 158.37500 217.66900 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 1 164.48200 217.66900 Tm (26) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 174.44500 217.66900 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 1 180.55200 217.66900 Tm (53) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 190.51400 217.66900 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 1 196.62100 217.66900 Tm (45) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 206.58400 217.66900 Tm [ (\135\054) -392.01300 (where) -363.00300 (the) -362.99100 (task) -364.00300 (is) ] TJ -156.47200 -11.95510 Td [ (to) -326.01200 (assign) -327.01800 (one) -326.01800 (label) -326.99300 (to) -326.01200 (an) -325.98400 (image\056) -540.00900 (Furt) 1.00351 (h) -1.01454 (e) 1.01454 (r) -327.00700 (w) 10 (ork) -326.01700 (has) -326.99700 (pushed) ] TJ 11.95510 TL T* [ (these) -219.01200 (adv) 24.98170 (ances) -218.00500 (along) -219.00400 (tw) 10.00810 (o) -219.01700 (orthogonal) -218.00500 (directions\072) -295.01200 (First\054) -224.98700 (rapid) ] TJ T* [ (progress) -227.98500 (in) -227.00900 (object) -227.98500 (detection) -228.01700 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 164.88100 181.80400 Tm (40) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 174.84400 181.80400 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 1 179.60600 181.80400 Tm (14) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 189.56800 181.80400 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 1 194.32100 181.80400 Tm (46) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 204.28300 181.80400 Tm [ (\135) -228.00400 (has) -227.99400 <6964656e7469026564> -227.99900 (mod\055) ] TJ -154.17100 -11.95630 Td [ (els) -211.99500 (that) -213.00800 (ef) 25.00810 <026369656e746c79> -212.00600 (identify) -212.99400 (and) -211.99800 (label) -213.01200 (multiple) -211.99600 (salient) -212.99900 (re) 15.00980 (gions) ] TJ 11.95510 TL T* [ (of) -295.00800 (an) -295.00900 (image\056) -445.00500 (Second\054) -306.98400 (recent) -295.01300 (adv) 24.98240 (ances) -295.01200 (in) -294.99700 (image) -294.99500 (captioning) ] TJ (\133) ' ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 53.42970 145.93800 Tm (3) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 58.41090 145.93800 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 64.37850 145.93800 Tm (32) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 74.34100 145.93800 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 80.30860 145.93800 Tm (21) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 90.27110 145.93800 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 96.23870 145.93800 Tm (49) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 106.20100 145.93800 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 1 112.16900 145.93800 Tm (51) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 122.13100 145.93800 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 1 128.08900 145.93800 Tm (8) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 133.07000 145.93800 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 1 139.03800 145.93800 Tm (4) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 144.01900 145.93800 Tm [ (\135) -349.00400 (ha) 19.99790 (v) 14.98280 (e) -349.00500 (e) 15.01220 (xpanded) -348.98300 (the) -348.99300 (comple) 14.99750 (xity) -349.01500 (of) ] TJ -93.90700 -11.95510 Td [ (the) -245.01000 (label) -245.00800 (space) -244.98100 (from) -245.00700 (a) -244.98400 <0278> 14.99750 (ed) -245.01700 (set) -245.00900 (of) -245.01700 (cate) 15.01100 (gories) -245.00300 (to) -245.00600 (sequence) -245.01300 (of) ] TJ 11.95510 TL T* [ (w) 10.00140 (ords) -249.98500 (able) -249.99400 (to) -249.98500 (e) 15.01220 (xpress) -249.98300 <7369676e690263616e746c79> -250.01600 (richer) -250 (concepts\056) ] TJ 11.95510 -11.95510 Td [ (Ho) 24.98420 (we) 25.01420 (v) 14.98280 (er) 39.98600 (\054) -380.01300 (despite) -355.01700 (encouraging) -353.99900 (progress) -354.00200 (along) -354.00200 (the) -354.99200 (label) ] TJ -11.95510 -11.95590 Td [ (density) -198.01400 (and) -198.98100 (label) -197.99500 (comple) 14.99870 (xity) -199.00100 (ax) 15.01830 (es\054) -208 (these) -199.01600 (tw) 10.00810 (o) -198.00100 (directions) -198.99400 (ha) 19.99670 (v) 14.98280 (e) ] TJ ET Q 3.98000 w 0 G 501.12100 904.14800 m 1446.11000 904.14800 l S q 10 0 0 10 0 0 cm BT /R46 5.97760 Tf 1 0 0 1 60.14100 83.81290 Tm [ <03> -0.90058 ] TJ /R36 7.97010 Tf 4.31680 -2.81289 Td [ (Both) -250.02000 (authors) -250.01700 (contrib) 19.99660 (uted) -250.01900 (equally) -250.01700 (to) -250.01300 (this) -250.00700 (w) 10.01210 (ork\056) ] TJ ET Q q 3088.62000 4494.57000 2362.50000 1287.74000 re W n /R64 Do Q q 10 0 0 10 0 0 cm BT /R36 8.96640 Tf 1 0 0 1 308.86200 438.49900 Tm [ (Figure) -390.00300 (1\056) -390.97800 (W) 80.01060 (e) -391.01900 (addres) 0.99656 (s) -390.99400 (the) -391.01100 (De) 1.01834 (nse) -391.01600 (Captioning) -390.01900 (task) -391.00500 (\050bottom) -389.98700 (right\051) ] TJ 10.95900 TL T* [ (with) -207.01500 (a) -208 (model) -206.98200 (that) -206.97900 (jointly) -207.98900 (generates) -207.01200 (both) -207.01700 (dense) -208.01400 (and) -207.00100 (rich) -207.00100 (annotations) ] TJ T* [ (in) -250.01400 (a) -249.99700 (single) -250.01900 (forw) 9.98193 (ard) -250.01100 (pass\056) ] TJ /R36 9.96260 Tf 22.88790 TL T* [ (remained) -245.98600 (separate\056) -309.00700 (In) -245.99600 (this) -246.01300 (w) 10.00320 (ork) -247.01300 (we) -245.99800 (tak) 10.00810 (e) -246.00300 (a) -246.00300 (step) -245.98300 (to) 24.98600 (w) 10.00320 (ards) -247.01300 (uni\055) ] TJ 11.95510 TL T* [ (fying) -294.99000 (these) -295 (tw) 10.00810 (o) -295.00500 (inter) 20.00890 (\055connect) 0.98513 (ed) -295.01000 (tasks) -294.99000 (into) -294.98500 (one) -295.00500 (joint) -295.01900 (frame\055) ] TJ 11.95590 TL T* [ (w) 10 (ork\056) -717.00800 (First\054) -418.99400 (we) -386.01600 (introduce) -385.00900 (the) -386.00400 (dense) -386.00400 (captioning) -384.99500 (task) -386 (\050see) ] TJ 11.95510 TL (Figure) ' ET Q 1 0 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 336.89700 357.82700 Tm (1) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 341.87800 357.82700 Tm [ (\051\054) -212.98400 (which) -203.99000 (requires) -203.01800 (a) -202.99100 (model) -204.02000 (to) -203.01500 (predict) -203.99000 (a) -202.99100 (set) -203.01500 (of) -204.00500 (descrip\055) ] TJ -33.01600 -11.95510 Td [ (tions) -270.98400 (across) -271.00900 (re) 15.00980 (gions) -270.00400 (of) -271.01100 (an) -271.01400 (image\056) -371.99700 (Object) -270.99900 (detection) -270.98900 (is) -270.99900 (hence) ] TJ 11.95470 TL T* [ (reco) 15.00980 (v) 14.98280 (ered) -363.00600 (as) -362.99600 (a) -363.98300 (special) -362.98600 (case) -363.00300 (when) -362.98800 (the) -362.98800 (tar) 17.99700 (get) -364.00800 (labels) -362.98400 (consist) ] TJ T* [ (of) -362.01600 (one) -362.01300 (w) 10 (ord\054) -390.00400 (and) -362.01300 (image) -362.00300 (captioning) -362.01800 (is) -362.98400 (reco) 15.01220 (v) 14.98280 (ered) -361.98400 (when) -362.00800 (all) ] TJ T* [ (images) -250.01500 (consist) -250.00500 (of) -249.99500 (one) -249.99300 (re) 15.00980 (gion) -250.01200 (that) -249.98300 (spans) -250.01700 (the) -249.98800 (full) -249.98300 (image\056) ] TJ 11.95510 -12.88520 Td [ (Additionally) 64.98920 (\054) -380.01300 (we) -355.00200 (de) 25.01540 (v) 14.98280 (elop) -354.00500 (a) -353.98500 (Fully) -353.98000 (Con) 40.01780 (v) 20.00160 (olutional) -354.99500 (Local\055) ] TJ -11.95510 -11.95470 Td [ (ization) -504.99200 (Netw) 10.00570 (ork) -503.98700 (\050FCLN\051) -504.98400 (for) -505.01400 (the) -503.98500 (dense) -505.00400 (captioning) -505.01400 (task\056) ] TJ T* [ (Our) -315.00100 (model) -316.00100 (is) -314.99400 (inspired) -315.01900 (by) -316.01600 (re) 1.00964 (cent) -315.98200 (w) 10.00320 (ork) -315.00100 (in) -314.99600 (image) -316.01100 (captioning) ] TJ (\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 261.25700 Tm (49) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 322.14200 261.25700 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 1 327.58200 261.25700 Tm (21) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 337.54400 261.25700 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 1 342.98400 261.25700 Tm (32) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 352.94600 261.25700 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 1 358.38600 261.25700 Tm (8) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 363.36700 261.25700 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 1 368.80600 261.25700 Tm (4) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 373.78800 261.25700 Tm [ (\135) -295.99200 (in) -296.01700 (that) -296.01400 (it) -296.01400 (is) -296.01400 (composed) -296 (of) -295.99000 (a) -295.99500 (Con) 40.01780 (v) 20.00160 (olutional) ] TJ -64.92540 -11.95510 Td [ (Neural) -265.01200 (Netw) 10.00570 (ork) -263.98800 (and) -265.01000 (a) -265.02000 (Recurrent) -263.98500 (Neural) -265.01000 (Netw) 10.00810 (ork) -265.01000 (language) ] TJ 11.95590 TL T* [ (model\056) -309.02000 (Ho) 24.98360 (we) 25.01300 (v) 14.98280 (er) 39.98600 (\054) -247.99500 (dra) 15.01470 (wing) -248.01700 (on) -247.00800 (w) 10.00320 (ork) -247.99300 (in) -247.98800 (object) -247.00300 (detection) -248.01300 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 529.34500 237.34600 Tm (38) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 539.30700 237.34600 Tm (\135\054) Tj -230.44500 -11.95510 Td [ (our) -231.99300 (second) -231.98800 (core) -232.00800 (contrib) 20.01870 (ution) -230.99100 (is) -231.98600 (to) -231.99100 (introduce) -232.01500 (a) -232.00500 (ne) 25.01540 (w) -231.99600 (dense) -231.99100 (lo\055) ] TJ 11.95510 TL T* [ (calization) -353.99700 (layer) 55.01040 (\056) -624.01100 (This) -353.99700 (layer) -355 (is) -354.00500 (fully) -355.01900 (dif) 24.98600 (ferentiable) -354.01000 (and) -354.99500 (can) ] TJ T* [ (be) -335.98300 (inserted) -336.00300 (into) -335.99800 (an) 15.01710 (y) -336.01700 (neural) -336.01700 (netw) 10.00810 (ork) -336.01700 (that) -336.00800 (processes) -336.01300 (images) ] TJ T* [ (to) -332.99100 (enable) -334.01800 (re) 15.00980 (gion\055le) 24.99830 (v) 14.98280 (el) -333 (training) -333.01300 (and) -334.01800 (predictions\056) -559.98500 (Internally) 64.98920 (\054) ] TJ 11.95630 TL T* [ (the) -279.00500 (localization) -279.01700 (layer) -279.01200 (predicts) -278.01200 (a) -279.01700 (set) -279.00200 (of) -279.01200 (re) 15.00730 (gions) -278.98300 (of) -279.01200 (interest) -278.99300 (in) ] TJ 11.95470 TL T* [ (the) -366.99000 (image) -366.00300 (and) -366.99300 (then) -366.98500 (uses) -366.00500 (bilinear) -367.01700 (interpolation) -366.00800 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 504.31800 165.61500 Tm (19) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 514.28100 165.61500 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 1 520.42800 165.61500 Tm (16) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R36 9.96260 Tf 1 0 0 1 530.39000 165.61500 Tm [ (\135) -367.00200 (to) ] TJ -221.52800 -11.95510 Td [ (smoothly) -249.98500 (crop) -249.99500 (the) -249.99000 (acti) 24.98110 (v) 24.98110 (ations) -250.01200 (in) -249.98800 (each) -250.00700 (re) 15.00730 (gion\056) ] TJ 11.95510 -12.88400 Td [ (W) 79.98660 (e) -216.00800 (e) 25.01050 (v) 24.98110 (aluate) -215.99600 (the) -215.99600 (model) -216.01800 (on) -215.99300 (the) -215.99300 (lar) 17.99700 (ge\055scale) -216.00300 (V) 59.99490 (isual) -215.98400 (Genome) ] TJ -11.95510 -11.95510 Td [ (dataset\054) -201.99100 (which) -189.99300 (contains) -189.98100 (94\054000) -190.01700 (images) -189.98800 (and) -190.98300 (4\054100\054000) -190.00800 (re) 15.00730 (gion) ] TJ 11.95590 TL T* [ (captions\056) -303.99800 (Our) -233.99500 (results) -232.99500 (sho) 24.99340 (w) -233.01500 (both) -234.01500 (performance) -232.98100 (and) -233.01500 (speed) -233.99000 (im\055) ] TJ 11.95510 TL T* [ (pro) 14.98520 (v) 14.98280 (ements) -280.99000 (o) 14.98280 (v) 14.98280 (er) -281.01900 (approaches) -281.00700 (based) -281.00200 (on) -281.00200 (pre) 25.01050 (vious) -282.00200 (state) -281.00200 (of) -281.01200 (the) ] TJ T* [ (art\056) -487.99500 (W) 79.98660 (e) -309.01200 (mak) 10.01060 (e) -309.01200 (our) -309 (code) -309.98500 (and) -309.00200 (data) -309.00200 (publicly) -309.00700 (a) 19.99180 (v) 24.98110 (ailable) -308.99300 (to) -310.01700 (sup\055) ] TJ T* [ (port) -249.98300 (further) -249.99500 (progress) -249.98000 (on) -249.98800 (the) -249.98800 (dense) -249.98800 (captioning) -249.99700 (task\056) ] 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 (4565) 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 1508
/Length 177372
/ColorSpace /DeviceRGB
/Width 2500
/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
$4%&'()*56789:CDEFGHIJSTUVWXYZcdefghijstuvwxyz ?
+{W1$Q3@
4 ;@"i@;
?ZZJ`-'ҔRڊ`h>{RR S֒9@r)hKH=:ҏJ;bԴb`Q@1>`?0LD/֫XqބF3Ҝu皰QE KIE K ){SRSҀ
(HB;&)GZ'ZjnuzU?Z*i)t[ZUE!5
~txSV( QILjTUz0dQ_ǧ784P!SSP\T2sUZ%WaLh4~Jn8RwB{RR