%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 (Ask Me Anything\072 Free\055Form Visual Question Answering Based on Knowledge From External Sources) /Producer (PyPDF2) /Author (Qi Wu\054 Peng Wang\054 Chunhua Shen\054 Anthony Dick\054 Anton van den Hengel) /Subject (2016 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 /x6 17 0 R /x12 20 0 R /x10 23 0 R /R40 26 0 R >> /ExtGState << /s9 29 0 R /s11 32 0 R /a0 << /CA 1 /ca 1 >> /R19 35 0 R /s5 36 0 R /s7 39 0 R >> /Font << /F2 42 0 R /R28 43 0 R /F1 47 0 R /R26 48 0 R /R37 52 0 R /R20 59 0 R /R22 62 0 R /R24 66 0 R /R33 70 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] /ColorSpace << /R31 27 0 R >> >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 ] >> endobj 4 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 87 0 R /Resources << /ExtGState << /R19 35 0 R >> /Font << /F2 88 0 R /R20 59 0 R /F1 89 0 R /R22 62 0 R /R28 43 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 90 0 R 91 0 R 92 0 R 93 0 R 94 0 R 95 0 R 96 0 R 97 0 R 98 0 R 99 0 R 100 0 R 101 0 R 102 0 R 103 0 R 104 0 R 105 0 R 106 0 R 107 0 R 108 0 R 109 0 R 110 0 R 111 0 R 112 0 R ] >> endobj 5 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 113 0 R /Resources << /XObject << /R83 114 0 R /R92 115 0 R >> /ExtGState << /R19 35 0 R >> /Font << /R93 116 0 R /R28 43 0 R /F2 119 0 R /R99 120 0 R /R89 123 0 R /F1 130 0 R /R20 59 0 R /R22 62 0 R /R95 131 0 R /R85 134 0 R /R97 141 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] /ColorSpace << /R31 27 0 R >> >> /MediaBox [ 0 0 612 792 ] /Annots [ 144 0 R 145 0 R 146 0 R 147 0 R 148 0 R 149 0 R 150 0 R ] >> endobj 6 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 151 0 R /Resources << /XObject << /R130 152 0 R /R131 153 0 R /R125 154 0 R >> /ExtGState << /R19 35 0 R >> /Font << /F2 155 0 R /R143 156 0 R /R127 163 0 R /R28 43 0 R /R132 170 0 R /F1 174 0 R /R134 175 0 R /R136 178 0 R /R20 59 0 R /R139 181 0 R /R22 62 0 R /R26 48 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] /ColorSpace << /R31 27 0 R >> >> /MediaBox [ 0 0 612 792 ] /Annots [ 188 0 R 189 0 R 190 0 R 191 0 R 192 0 R 193 0 R 194 0 R 195 0 R 196 0 R 197 0 R 198 0 R 199 0 R 200 0 R 201 0 R 202 0 R 203 0 R 204 0 R ] >> endobj 7 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 205 0 R /Resources << /ExtGState << /R19 35 0 R >> /Font << /R163 206 0 R /R161 209 0 R /R167 213 0 R /R165 217 0 R /R175 221 0 R /R169 224 0 R /R171 228 0 R /R173 232 0 R /F2 235 0 R /R28 43 0 R /R132 170 0 R /F1 236 0 R /R134 175 0 R /R136 178 0 R /R20 59 0 R /R22 62 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 << /R19 35 0 R >> /Font << /F2 245 0 R /R28 43 0 R /R132 170 0 R /F1 246 0 R /R134 175 0 R /R20 59 0 R /R22 62 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 247 0 R 248 0 R 249 0 R 250 0 R 251 0 R 252 0 R 253 0 R 254 0 R 255 0 R 256 0 R 257 0 R 258 0 R 259 0 R 260 0 R 261 0 R 262 0 R 263 0 R 264 0 R ] >> endobj 9 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 265 0 R /Resources << /ColorSpace << /R31 27 0 R >> /ExtGState << /R19 35 0 R >> /Font << /R163 206 0 R /R161 209 0 R /R215 266 0 R /R219 273 0 R /F2 280 0 R /R28 43 0 R /R132 170 0 R /F1 281 0 R /R136 178 0 R /R20 59 0 R /R22 62 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 282 0 R 283 0 R 284 0 R 285 0 R 286 0 R 287 0 R 288 0 R 289 0 R 290 0 R 291 0 R ] >> endobj 10 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 292 0 R /Resources << /XObject << /R246 293 0 R /R247 294 0 R /R244 295 0 R /R245 296 0 R /R242 297 0 R /R243 298 0 R /R240 299 0 R /R241 300 0 R >> /ExtGState << /R19 35 0 R >> /Font << /F2 301 0 R /R20 59 0 R /F1 302 0 R /R22 62 0 R /R28 43 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 303 0 R 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 ] >> endobj 11 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 316 0 R /Resources << /ExtGState << /R19 35 0 R >> /Font << /F2 317 0 R /R28 43 0 R /F1 318 0 R /R22 62 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 319 0 R 320 0 R 321 0 R 322 0 R 323 0 R 324 0 R 325 0 R 326 0 R 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 340 0 R 341 0 R 342 0 R 343 0 R 344 0 R 345 0 R 346 0 R 347 0 R 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 358 0 R 359 0 R 360 0 R 361 0 R 362 0 R 363 0 R 364 0 R 365 0 R 366 0 R ] >> endobj 12 0 obj << /Type /Catalog /Pages 1 0 R >> endobj 13 0 obj << /Length 20733 >> stream q q q 0.10000 0 0 0.10000 0 0 cm /R19 gs 0 g q 10 0 0 10 0 0 cm BT /R20 14.34620 Tf 1 0 0 1 122.42500 675.06700 Tm [ (Ask) -249.99300 (Me) -250.01000 (Anything\072) -310.00500 (Fr) 18.00480 (ee\055f) 25.01270 (orm) -249.99600 (V) 37.01370 (isual) -250.00500 (Question) -249.99600 (Answering) ] TJ 40.73590 -17.93320 Td [ (Based) -250.00200 (on) -250.01300 (Kno) 9.99285 (wledge) -250.00800 (fr) 18.00650 (om) -250.00800 (Exter) 15.01140 (nal) -249.98800 (Sour) 18.00480 (ces) ] TJ /R22 11.95520 Tf -43.33910 -37.85780 Td [ (Qi) -250.01200 (W) 50.00460 (u\054) -249.98900 (Peng) -250.00200 (W) 80 (ang\054) -250.01000 (Chunhua) -249.99500 (Shen\054) -249.99100 (Anthon) 14.99540 (y) -250.00200 (Dick\054) -249.98700 (Anton) -250.01400 (v) 24.98340 (an) -250.01000 (den) -249.98500 (Hengel) ] TJ 15.99610 -13.94730 Td [ (School) -250.00400 (of) -250.01500 (Computer) -250.01400 (Science\054) -249.99300 (The) -250.01400 (Uni) 24.99570 (v) 14.98510 (ersity) -249.98900 (of) -250.01400 (Adelaide\054) -250.00600 (Australia) ] TJ /R24 8.96640 Tf -44.55700 -13.94800 Td [ (f) -0.89990 ] TJ /R26 8.96640 Tf 4.60820 0 Td (qi\056wu01\054p\056wang\054chunhua\056shen\054anthony\056dick\054anton\056vandenhengel) Tj /R24 8.96640 Tf 317.40900 0 Td [ (g) -0.89854 ] TJ /R26 8.96640 Tf 4.60781 0 Td (\100adelaide\056edu\056au) Tj /R20 11.95520 Tf -271.89100 -41.04570 Td (Abstract) Tj /R28 9.96260 Tf -83.92770 -25.25510 Td [ (W) 91.98650 (e) -513.99700 (pr) 44.98390 (opose) -512.99500 (a) -513.98800 (method) -514.01100 (for) -512.99400 (visual) -514.00600 (question) -513.99500 (answering) ] TJ -11.95510 -11.95510 Td [ (whic) 14.99870 (h) -239.01400 (combines) -240.01600 (an) -239.00900 (internal) -238.98100 (r) 37.01960 (epr) 36.98160 (esentation) -240 (of) -239.00700 (the) -239.01200 (content) -240.01900 (of) ] TJ 11.95510 TL T* [ (an) -257.98600 (ima) 10.01360 (g) 10.00320 (e) -256.98100 (with) -258.00200 (information) -258.01200 (e) 19.99180 (xtr) 14.98650 (acted) -257.98800 (fr) 44.98640 (om) -257.00900 (a) -257.99100 (g) 10.00320 (ener) 15.01960 (al) -257.98400 (knowl\055) ] TJ T* [ (edg) 10.00380 (e) -349.98500 (base) -350.00800 (to) -350.00800 (answer) -349.98900 (a) -350.01500 (br) 44.98390 (oad) -350.00500 (r) 14.98400 (ang) 10.01300 (e) -349.98600 (of) -350.00800 (ima) 10.01300 (g) 10.00320 (e\055based) -350.01000 (ques\055) ] TJ T* [ (tions\056) -458.98800 (This) -299.99600 (allows) -299.98400 (mor) 36.98770 (e) -299.99400 (comple) 20.00770 (x) -299.01400 (questions) -299.98700 (to) -300.01600 (be) -299.98900 (answer) 37.00610 (ed) ] TJ T* [ (using) -202.99700 (the) -201.99800 (pr) 36.98520 (edominant) -202.98700 (neur) 14.99010 (al) -201.99300 (network\055based) -202.98300 (appr) 44.99490 (oac) 14.98280 (h) -202.98100 (than) ] TJ 11.95590 TL T* [ (has) -342.98600 (pr) 36.98520 (e) 15.01280 (viously) -342 (been) -343 (possible) 15.01470 (\056) -586.99800 (It) -342.99800 (particularly) -342.00900 (allows) -342.99700 (ques\055) ] TJ 11.95510 TL T* [ (tions) -234.00900 (to) -233.98800 (be) -234 (ask) 10.00570 (ed) -233.99900 (about) -234.01200 (the) -233.99300 (contents) -234.01200 (of) -233.98800 (an) -233.99000 (ima) 10.01300 (g) 10.00320 (e) 9.99343 (\054) -237.01700 (e) 15.01220 (ven) -234.00300 (when) ] TJ T* [ (the) -392.00500 (ima) 10.01360 (g) 10.00320 (e) -392.01700 (itself) -391.98200 (does) -392 (not) -391.99600 (contain) -391.98200 (the) -392.00600 (whole) -391.98600 (answer) 110.99400 (\056) -737.01200 (The) ] TJ T* [ (method) -306.00700 (constructs) -305.98800 (a) -305.98300 (te) 20 (xtual) -306.00800 (r) 37.01960 (epr) 36.98160 (esentation) -305.98800 (of) -306.99500 (the) -306.02000 (semantic) ] TJ T* [ (content) -385.99300 (of) -384.98200 (an) -386.00400 (ima) 10.01300 (g) 10.00320 (e) 9.99343 (\054) -420.00600 (and) -386 (mer) 37.01830 (g) 10.00320 (es) -384.99300 (it) -386 (with) -385.98200 (te) 20 (xtual) -385.99400 (informa\055) ] TJ T* [ (tion) -357.99400 (sour) 36.99630 (ced) -357.98300 (fr) 44.98640 (om) -358.99200 (a) -358.01400 (knowledg) 9.99098 (e) -357.98400 (base) 10.01060 (\054) -384.99200 (to) -358.00600 (de) 15.01710 (velop) -358.99200 (a) -358.01400 (deeper) ] TJ 11.95630 TL T* [ (under) 10.01490 (standing) -408.00400 (of) -407.99800 (the) -408.00300 (scene) -408.01300 (vie) 15.01590 (wed\056) -784.01000 (Priming) -408.01300 (a) -408.00500 (r) 37.01830 (ecurr) 36.98160 (ent) ] TJ 11.95470 TL T* [ (neur) 14.98950 (al) -441.99200 (network) -442.99600 (with) -442.01100 (this) -442.01800 (combined) -443.00300 (information\054) -490.00900 (and) -443.00900 (the) ] TJ T* [ (submitted) -301.98700 (question\054) -315.00600 (leads) -302.01000 (to) -302.01600 (a) -301.98400 (very) -303.01600 <0365> 19.99790 (xible) -302.01800 (visual) -302.00100 (question) ] TJ T* [ (answering) -398.98300 (appr) 44.99370 (oac) 14.98400 (h\056) -759.00300 (W) 91.98590 (e) -398.99700 (ar) 36.98520 (e) -398.99700 <73706563690263616c6c79> -399.98700 (able) -399.01900 (to) -400 (answer) ] TJ T* [ (questions) -275.99000 (posed) -274.99600 (in) -275.98100 (natur) 15.00610 (al) -275.98100 (langua) 9.99098 (g) 10.00320 (e) 9.99343 (\054) -281.99000 (that) -274.98800 (r) 37.01830 (efer) -275.98800 (to) -275.98100 (informa\055) ] TJ T* [ (tion) -288.98600 (not) -288.99300 (contained) -289.01900 (in) -288.99800 (the) -287.98400 (ima) 10.01300 (g) 10.00320 (e) 15.01220 (\056) -426.98500 (W) 91.98590 (e) -289.01600 (demonstr) 15.00980 (ate) -289.00300 (the) -289.00300 (ef) 18 (fec\055) ] TJ 11.95590 TL T* [ (tiveness) -377.01700 (of) -378.00300 (our) -377.01900 (model) -378 (on) -376.98600 (two) -377.99000 (publicly) -377.00800 (available) -377.98300 (datasets\054) ] TJ 11.95510 TL T* [ (T) 91.99870 (or) 44.98390 (onto) -234.01700 (COCO\055QA) -234.00400 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R28 9.96260 Tf 1 0 0 1 133.66900 297.93200 Tm (23) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R28 9.96260 Tf 1 0 0 1 143.63100 297.93200 Tm [ (\135) -233.99400 (and) -233.00500 (VQA) -233.98900 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R28 9.96260 Tf 1 0 0 1 192.67700 297.93200 Tm (1) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R28 9.96260 Tf 1 0 0 1 197.65800 297.93200 Tm [ (\135) -233.99400 (and) -233.98500 (show) -234.01000 (that) -234.01500 (it) -233.98500 (pr) 44.98510 (o\055) ] TJ -147.54600 -11.95470 Td [ (duces) -249.99000 (the) -249.99000 (best) -249.98400 (r) 37.01960 (eported) -250.01700 (r) 37.01960 (esults) -250.01000 (in) -249.98500 (both) -250.01500 (cases\056) ] TJ /R20 11.95520 Tf 39.89800 TL T* [ (1\056) -249.99000 (Intr) 18.01460 (oduction) ] TJ /R22 9.96260 Tf 11.95510 -20.27300 Td [ (V) 59.99310 (isual) -266.99300 (question) -267.01800 (answering) -266.99500 (\050VQA\051) -267.01200 (is) -267.00200 (distinct) -267.01200 (from) -267.00400 (man) 14.99010 (y) ] TJ -11.95510 -11.95510 Td [ (problems) -389 (in) -388.00100 (Computer) -388.99300 (V) 59.99240 (ision) -387.98300 (because) -388.99600 (the) -388.98700 (quest) 1 (ion) -389.01600 (to) -388.98200 (be) ] TJ 11.95510 TL T* [ (answered) -275.01000 (is) -274.01900 (not) -274.99600 (determined) -274.98700 (until) -274.98100 (run) -273.98600 (time) -274.99100 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R22 9.96260 Tf 1 0 0 1 223.67000 201.89600 Tm (1) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R22 9.96260 Tf 1 0 0 1 228.65200 201.89600 Tm [ (\135\056) -384.01400 (In) -275.01000 (more) -275.00300 (tra\055) ] TJ -178.53900 -11.95590 Td [ (ditional) -420.00800 (problems) -418.99500 (such) -419.99600 (as) -420.00700 (se) 15.01830 (gmentation) -420.00300 (or) -420.00600 (dete) 0.98268 (ction\054) -462.99900 (the) ] TJ 11.95510 TL T* [ (single) -375.98900 (question) -376.01900 (to) -376.00300 (be) -376.01400 (answered) -376.01300 (by) -376.00600 (an) -376.01400 (algorithm) -376.01300 (is) -376.00300 (prede\055) ] TJ T* [ (termined\054) -357.01200 (and) -336.01600 (only) -336 (the) -334.99600 (image) -336.00600 (changes\056) -566.99400 (In) -335.98100 (visual) -335.99500 (question) ] TJ T* [ (answering\054) -304.98600 (in) -294.99700 (contrast\054) -304.98700 (the) -293.98300 (form) -294.99900 (that) -294.01500 (the) -293.98300 (question) -293.99300 (will) -294.98500 (tak) 10.00570 (e) ] TJ T* [ (is) -229.00700 (unkno) 25.00130 (wn\054) -233.01000 (as) -228 (is) -229.00700 (the) -229.01300 (set) -227.99300 (of) -229.01900 (operations) -228.99200 (required) -227.99400 (to) -229.00800 (answer) -229.01800 (it\056) ] TJ T* [ (In) -210.98300 (this) -210.99800 (sense) -209.99600 (it) -211.00900 (more) -211.01400 (closely) -211.00200 <72650365637473> -209.99300 (the) -211.01600 (challenge) -211.01100 (of) -210.98200 (general) ] TJ 11.95590 TL T* [ (image) -249.98200 (interpretation\056) ] TJ 11.95510 -13.29920 Td [ (VQA) -281.00300 (typically) -279.99600 (requires) -281.00300 (processing) -280.00500 (both) -280.99000 (visual) -280.98500 (informa\055) ] TJ -11.95510 -11.95510 Td [ (tion) -320.98100 (\050the) -319.98200 (image\051) -320.99300 (and) -320.01900 (te) 14.98030 (xtual) -320.98600 (information) -321 (\050the) -319.98100 (question) -321.00800 (and) ] TJ 11.95510 TL T* [ (answer\051\056) -702 (One) -380.01100 (approach) -380.99200 (to) -380.00300 (V) 59.99240 (ision\055to\055Language) -380.98800 (problems\054) ] TJ ET Q q 3088.62000 3231.15000 2362.47000 2371.84000 re W n /R31 cs 1 1 1 scn 3088.62000 3231.15000 2362.47000 2371.84000 re f 0.12207 0.28589 0.48999 scn q 10 0 0 10 0 0 cm BT /R33 9.71200 Tf 1 0 0 1 310.97000 427.02400 Tm [ <00010002> 8.81843 <0003> 9.66809 <00040005> -9.48207 <00020006000700080009> 82.33460 <0004> 9.00696 <000a0003000b> -6.38506 <000600070008000c> 10.92750 <0004000d> -8.19500 <00050004000e00040002> 7.85313 <0003> 9.66809 <0006> 12.07130 <0003000f00100002> -6.70683 <00110008> ] TJ ET Q 0 0 0 scn q 10 0 0 10 0 0 cm BT /R37 8.77590 Tf 1 0 0 1 310.96800 416.25400 Tm [ <0001> -487.66400 <00020003> 20.01890 <000400050006> -489.67500 <00040007> -514.67100 <00060008> 9.67004 <000400060009> -12.33790 <0008> -501.98500 <0008000a> 9.65056 <000b0004> -7.31095 <000c000d000a> 6.97712 <0002> -489.01600 <000e> -507.67700 <000f0005> 9.33620 <000a> -8.33470 <000a> 18.32740 <000c> -493.66200 <0010000e> 17.33150 <000c> -493.66200 <000e> 12.34630 <0011> -491.65900 <0011> -11.65080 <00120008> -497 <00130008> 9.67004 <000e0014> -18.01590 <0012> -488.34200 <0015000d00110012> ] TJ ET Q 4297.36000 4146.18000 210.62100 5.85156 re f 4998.26000 4146.18000 215.30100 5.85156 re f q 10 0 0 10 0 0 cm BT /R37 8.77590 Tf 1 0 0 1 310.96800 405.72300 Tm [ <000500160013> 8.98290 <0003> 15.65680 <00080009> 6.97990 <0009000e000f> -247.64600 <000d000a> -232.66800 <001100120008> -241.99600 <000f000e000a> 8.31523 <00100017> ] TJ ET Q 3110.85000 4040.88000 349.86700 5.84766 re f 0.75391 0.31396 0.30200 scn q 10 0 0 10 0 0 cm BT /R33 9.71200 Tf 1 0 0 1 310.97000 394.25500 Tm [ <0012> 6.07335 <000a0003> 10.83450 <00040005> -9.48207 <0002000600070008> -15.08280 <0013> 16.88770 <00020010> -9.32622 <00140007000400150016> 11.01800 <000400110008> ] TJ ET Q 0 0 0 scn q 10 0 0 10 0 0 cm BT /R37 8.77590 Tf 1 0 0 1 310.96800 383.49300 Tm [ <0001000a> -469.33300 <000500160013> 8.98290 <00030008> 7 <00090009000e> -476.34100 <000d000f> -460.00800 <000e> -481.01500 <0014> 9.67004 <000e000a00040006000c> -477.68200 <00100008> 9.67004 <000f000d> -6.68779 <0002> -9.00793 <000a00080010> -465.33000 <0011> 15.01140 <0004> -486.33900 <00060003> 20.68650 <00040011> 8.69079 <000800140011> -464.01700 <000e0002> 16.67500 <000e000d> -11.98460 <000a000f> 22.67280 <0011> -478.35000 <0003> 29.01010 <000e000d> -11.98460 <000a> -461.68000 <00040003> ] TJ ET Q 3248.93000 3818.55000 317.10200 5.85156 re f q 10 0 0 10 0 0 cm BT /R37 8.77590 Tf 1 0 0 1 310.96800 372.96200 Tm [ <000f0005> 9.33899 <000a> -8.33192 <0009000d00020012> 14.01270 <00110017> -239.67000 <0018000e> -7.68373 <0003> 15.65680 <00020008> 9.00237 <0003> -237.65600 <0005001600130003> 11.33920 <00080009> 6.97990 <0009000e000f> -234.33700 <000e00030008> -233.98400 <00040007> -8 <0011> 15.01140 <0008000a> -230.35600 <0005> -8.33470 <000f00080010> -239.34700 <000e000f> -236.64900 <00060004000d> -12.34070 <000a> 18.32740 <0011000f> -233.97800 <00040007> -248.00500 <000f0012> 9.33620 <000e0010> -9.31395 <0008> -235.31900 <0004000a> -241.35000 <000e> ] TJ ET Q 3680.70000 3713.24000 351.03900 5.85156 re f 5027.51000 3713.24000 211.79300 5.85156 re f q 10 0 0 10 0 0 cm BT /R37 8.77590 Tf 1 0 0 1 310.96800 362.43100 Tm [ <000f0005> 9.33899 <000a> -8.33192 <000a> 18.33020 <000c> -253.65800 <00130008> 9.67004 <000e0014> -18.01590 <00120017> -249.68500 <0001> -247.66000 <0013> -8.33192 <0008000e00140012> -248.33600 <000d000f> -246.66700 <000e> -254.32000 <0009000e> -12.00130 <000a0010> 10.00390 <0007> 24.98190 <0004> -19.68500 <00030016> -251.99400 <000e00090004000a0002> -249.01200 <00110012> -6.68779 <0008> -248.67200 <0014> 9.67004 <0004000e> -7.32764 <000f> 17.68760 <0011> -251.65500 <00040007> -261.31400 <000e000a> -249.31800 <00040014> -10.00390 <0008000e000a> 8.66854 <0017> ] TJ ET Q 3110.85000 3607.93000 448.15600 5.84766 re f 3676.02000 3607.93000 215.30500 5.84766 re f q 10 0 0 10 0 0 cm BT /R37 8.77590 Tf 1 0 0 1 310.96800 351.90000 Tm [ <00190011> -226.33100 <0005000f> 9.31673 <0005000e00090009> 8.67132 <000c> -253.65800 <0014> 9.67004 <0004000a000f000d000f> 22.98160 <0011000f> -233.98100 <00040007> -234.65100 <000900040004> -10.33770 <000f0008> -217.64800 <0006000e00030011> 8.02313 <000d001400090008000f> 9.98718 <001a> -243.31400 <000f0005> 9.33620 <00140012> -238.68000 <000e000f> -236.64900 <000f000e000a> 8.31245 <0010001b0017> ] TJ ET Q 4681.15000 3502.62000 167.32800 5.85156 re f 0.50195 0.39209 0.63477 scn q 10 0 0 10 0 0 cm BT /R33 9.71200 Tf 1 0 0 1 310.97000 340.43200 Tm [ <0017000b0004> -8.93155 <000e> 13.44380 <0003000f00100002> -235.60400 <00180002000e0014> 9.88428 <00040005> -9.48207 <000f000200160011> ] TJ 11.70310 TL <00170011> ' ET Q 0 0 0 scn q 10 0 0 10 0 0 cm BT /R37 8.77590 Tf 1 0 0 1 322.55200 328.73200 Tm [ <001c0012> 14.99190 <000c> -226.99600 <00100004> -227.99700 <001100120008> 11.31690 <000c> -240.34900 <0012000e> 17.33150 <001d> 12.00690 <0008> -235.31900 <000500160013> 8.98290 <0003> 15.65680 <00080009> 6.97712 <0009000e000f001e> ] TJ ET Q 0.50195 0.39209 0.63477 scn q 10 0 0 10 0 0 cm BT /R33 9.71200 Tf 1 0 0 1 432.19500 328.72900 Tm [ <0018> -237.39400 <0011> ] TJ ET Q 0 0 0 scn q 10 0 0 10 0 0 cm BT /R37 8.77590 Tf 1 0 0 1 445.17800 328.73200 Tm [ <001f0012> 10.67150 <000e00100008> 8.64628 <0017> ] TJ ET Q 0 0.68945 0.31396 scn q 10 0 0 10 0 0 cm BT /R33 9.71200 Tf 1 0 0 1 497.37200 545.19900 Tm [ <0018> 39.72820 <0003> 9.66809 <00030005> -8.83854 <000f0019000b00030004000e0011> ] TJ ET Q 0 0 0 scn q 10 0 0 10 0 0 cm BT /R37 8.77590 Tf 1 0 0 1 497.36800 534.43900 Tm [ <000500160013> 8.98011 <0003> 15.65680 <00080009> 6.97712 <0009000e> ] TJ 10.53130 TL T* [ <00130008> 9.67004 <000e00140012> ] TJ T* [ <000f0005> 9.33620 <000a000a> 23.31270 <000c> ] TJ T* [ <0010000e> 17.33150 <000c> ] TJ T* [ <00060008> 9.67004 <0004000600090008> ] TJ T* [ <000f000e000a> 8.31245 <0010> ] TJ T* [ <0009000e> 14.66080 <000c000d000a> 6.97712 <0002> ] TJ T* [ <00130009> 7.33320 <00050008> ] TJ T* [ <00020003> 20.01890 <00080008> 9.31395 <000a> ] TJ T* [ <00160004> -7.31095 <0005000a> 23.31270 <0011> 15.01140 <000e000d000a> ] TJ ET Q q 1827.73000 0 0 1190.01000 3113.19000 4365 cm /R40 Do Q Q q 10 0 0 10 0 0 cm BT /R22 8.96640 Tf 1 0 0 1 308.86200 312.15700 Tm [ (Figure) -395.01300 (1\056) -394.98600 (A) -395.01900 (real) -395.00800 (case) -396.01800 (of) -394.97800 (question) -395.00800 (answering) -395.01300 (based) -394.99700 (on) -395.00800 (an) -394.98600 (inter) 19.98020 (\055) ] TJ 10.95900 TL T* [ (nal) -322.00300 (te) 15.01920 (xtual) -321.99200 (representation) -322.02200 (and) -322.01400 (e) 15.01370 (xternal) -322.98300 (kno) 25.00660 (wledge\056) -525.98200 (All) -321.98100 (of) -322.00800 (the) -322.00300 (at\055) ] TJ T* [ (trib) 20 (utes\054) -453.02000 (te) 15.01920 (xtual) -413 (representation\054) -452.99900 (kno) 25.00660 (wledge) -412.99500 (and) -413.02200 (answer) -413.01700 (are) -412.00400 (pro\055) ] TJ T* [ (duced) -324.99300 (by) -324.99800 (our) -325.01200 (VQA) -323.98200 (model\056) -535.00800 (Underlined) -324.99800 (w) 10.00920 (ords) -325.00900 (indicate) -324.98200 (the) -324.00700 (infor) 20.01830 (\055) ] TJ T* [ (mation) -250.01400 (required) -249.99200 (to) -250.01400 (answer) -249.99200 (the) -249.98900 (question\056) ] TJ /R22 9.96260 Tf 31.87890 TL T* [ (such) -360.98600 (as) -360.01600 (VQA) -360.99100 (and) -360.99400 (i) 0.98758 (mage) -360.98900 (captioning\054) -388.01400 (which) -360.98400 (interrelate) -360.98900 (vi\055) ] TJ 11.95510 TL T* [ (sual) -395.99800 (and) -396.00700 (te) 14.98030 (xtual) -395.99300 (information) -396.00700 (is) -395.99800 (based) -396.00300 (on) -396.98300 (a) -396.01700 (direct) -396.00300 (method) ] TJ T* [ (pioneered) -265.98500 (in) -265.98300 (machine) -264.99700 (language) -265.98000 (translation) -265.99500 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R22 9.96260 Tf 1 0 0 1 484.23300 212.53200 Tm (7) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R22 9.96260 Tf 1 0 0 1 489.21400 212.53200 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R22 9.96260 Tf 1 0 0 1 494.35500 212.53200 Tm (26) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R22 9.96260 Tf 1 0 0 1 504.31800 212.53200 Tm [ (\135\056) -357 (This) -266.01000 (di\055) ] TJ -195.45500 -11.95510 Td [ (rect) -295.99500 (approach) -295.00500 (de) 25.01540 (v) 14.98280 (elops) -296.00900 (an) -295.99000 (encoding) -296.00500 (of) -295.01000 (the) -295.98000 (input) -296 (te) 14.98280 (xt) -296.01900 (using) ] TJ 11.95590 TL T* [ (a) -222.98700 (Recurrent) -223.01200 (Neural) -223.01900 (Netw) 10.00570 (ork) -223.01700 (\050RNN\051) -223.00700 (and) -223.01700 (passes) -223.98700 (it) -223.00700 (to) -223.01200 (another) ] TJ 11.95510 TL T* [ (RNN) -250.01700 (for) -250 (decoding\056) ] TJ 11.95510 -11.98010 Td [ (The) -414.00400 (V) 59.99240 (ision\055to\055Language) -412.98000 (v) 14.98280 (ersion) -413.99500 (of) -414.00900 (thi) 1 (s) -414.00400 (direct) -414 (transla\055) ] TJ -11.95510 -11.95470 Td [ (tion) -240.01400 (approach) -239.01400 (uses) -239.98700 (a) -238.98500 (Con) 40.01540 (v) 20.00160 (olutional) -239.99400 (Neural) -239.01400 (Netw) 10.00810 (ork) -239.99400 (\050CNN\051) ] TJ T* [ (to) -405.99800 (encode) -406.00800 (the) -406.98400 (image\054) -445.00700 (and) -406.00600 (an) -406.01100 (RNN) -405.99100 (to) -406 (encode) -406.99100 (a) 1.01454 (n) -1.01454 (d) -406.00600 (gener) 19.99670 (\055) ] TJ T* [ (ate) -402.01400 (te) 14.98030 (xt\056) -764.99500 (This) -401.98900 (approach) -402.00600 (has) -402.00600 (been) -402.01100 (well) -401.01600 (studied\056) -765.99500 (The) -402.00600 (im\055) ] TJ T* [ (age) -382.99700 (captioning) -384.01700 (methods) -382.99700 (de) 25.01540 (v) 14.98280 (eloped) -384 (in) -382.98500 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R22 9.96260 Tf 1 0 0 1 469.23000 116.86600 Tm (9) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R22 9.96260 Tf 1 0 0 1 474.21100 116.86600 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R22 9.96260 Tf 1 0 0 1 480.51800 116.86600 Tm (14) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R22 9.96260 Tf 1 0 0 1 490.48000 116.86600 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R22 9.96260 Tf 1 0 0 1 496.79600 116.86600 Tm (21) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R22 9.96260 Tf 1 0 0 1 506.75900 116.86600 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R22 9.96260 Tf 1 0 0 1 513.06500 116.86600 Tm (28) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R22 9.96260 Tf 1 0 0 1 523.02800 116.86600 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R22 9.96260 Tf 1 0 0 1 529.34400 116.86600 Tm (32) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R22 9.96260 Tf 1 0 0 1 539.30700 116.86600 Tm (\135\054) Tj -230.44500 -11.95590 Td [ (for) -414.99200 (e) 15.01220 (xample\054) -456.02000 (learn) -414.99000 (a) -414.99400 (mapping) -414.01400 (from) -415.01900 (images) -415.00900 (to) -415.01900 (te) 14.98280 (xt) -415.01900 (using) ] TJ 11.95510 TL T* [ (CNNs) -266.01000 (trained) -265.98300 (on) -265.98500 (object) -266.01900 (recognition\054) -269.99400 (and) -265.99000 (w) 10.00320 (ord) -265.99000 (embeddings) ] TJ T* [ (trained) -251.98500 (on) -253.00700 (lar) 17.99700 (ge) -251.99700 (scale) -251.99700 (te) 14.98030 (xt) -251.98700 (corpora\056) -317.00600 (V) 59.99490 (isual) -252.01700 (question) -252.98200 (answer) 19.99670 (\055) ] 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 32 72 14 re f* 0.50000 0.50000 0.50000 rg BT /F2 9 Tf 10.80000 TL ET BT 1 0 0 1 297 35 Tm (4622) 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 /FlateDecode
/Resources <<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
/XObject <<
/x15 18 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
18 0 obj
<<
/Filter /FlateDecode
/Resources 19 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
19 0 obj
<<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
>>
endobj
20 0 obj
<<
/Filter /FlateDecode
/Resources <<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
/XObject <<
/x24 21 0 R
>>
>>
/Length 28
/Group <<
/Type /Group
/S /Transparency
/CS /DeviceRGB
/I true
>>
/BBox [ 132 751 480 772 ]
/Type /XObject
/Subtype /Form
>>
stream
x+O4PH/VЯ02Qp
0
endstream
endobj
21 0 obj
<<
/Filter /FlateDecode
/Resources 22 0 R
/Length 53223
/Type /XObject
/BBox [ 132 751 480 772 ]
/Subtype /Form
>>
stream
xtI:6%Q㨈?7rA= u%6 ?Y(WbWo{B>9
x`Znϳ|8{3?0x*z ǃ|,@:w>`c|*ϻⳅKO3`g
:_|}}><.6`Z{{3]#<_o"~:ͺgk7/Ұ@|K yp ]03ʷCmş8˽Y?>(3!Bwqs.Z8,~~=rMT̩y+/*w: uBZ_`ߵp`%M?ɝ1ɳw=vDۉy&xb4Q>d@ sg~lA