%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 (Comprehension\055Guided Referring Expressions) /Producer (PyPDF2) /Author (Ruotian Luo\054 Gregory Shakhnarovich) /Subject (2017 IEEE Conference on Computer Vision and Pattern Recognition) >> endobj 3 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 14 0 R /Resources << /XObject << /x8 15 0 R /R32 18 0 R /x6 19 0 R /R30 22 0 R /x12 23 0 R /R33 26 0 R /x10 29 0 R /R31 32 0 R >> /ExtGState << /s9 33 0 R /R12 36 0 R /s11 37 0 R /a0 << /CA 1 /ca 1 >> /s5 40 0 R /s7 43 0 R >> /Font << /F2 46 0 R /R13 47 0 R /R15 51 0 R /R17 55 0 R /R19 58 0 R /R23 62 0 R /R25 66 0 R /F1 70 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] /ColorSpace << /R29 27 0 R /R22 71 0 R >> >> /MediaBox [ 0 0 612 792 ] /Annots [ 73 0 R 74 0 R 75 0 R 76 0 R ] >> endobj 4 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 77 0 R /Resources << /XObject << /R72 78 0 R /R73 79 0 R /R78 80 0 R >> /ExtGState << /R12 36 0 R >> /Font << /F2 81 0 R /R79 82 0 R /R15 51 0 R /R13 47 0 R /R70 86 0 R /R76 90 0 R /R81 94 0 R /R74 99 0 R /F1 104 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] /ColorSpace << /R29 27 0 R /R22 71 0 R >> >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 125 0 R 126 0 R 127 0 R 128 0 R 129 0 R 130 0 R 131 0 R 132 0 R ] >> endobj 5 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 133 0 R /Resources << /ExtGState << /R12 36 0 R >> /Font << /R108 134 0 R /R79 82 0 R /R112 137 0 R /F1 140 0 R /R110 141 0 R /R116 145 0 R /R104 149 0 R /R74 99 0 R /R106 153 0 R /F2 158 0 R /R13 47 0 R /R114 159 0 R /R17 55 0 R /R81 94 0 R /R15 51 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 6 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 176 0 R /Resources << /XObject << /R78 80 0 R >> /ExtGState << /R12 36 0 R /R133 177 0 R /R134 178 0 R >> /Font << /F2 179 0 R /R79 82 0 R /R74 99 0 R /F1 180 0 R /R13 47 0 R /R81 94 0 R /R104 149 0 R /R114 159 0 R /R15 51 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] /ColorSpace << /R29 27 0 R /R22 71 0 R >> >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 ] >> endobj 7 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 191 0 R /Resources << /ExtGState << /R12 36 0 R >> /Font << /R108 134 0 R /R79 82 0 R /R112 137 0 R /F1 192 0 R /R104 149 0 R /R74 99 0 R /R106 153 0 R /F2 193 0 R /R13 47 0 R /R15 51 0 R /R19 58 0 R /R81 94 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 194 0 R 195 0 R 196 0 R 197 0 R 198 0 R ] >> endobj 8 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 199 0 R /Resources << /ExtGState << /R12 36 0 R >> /Font << /R79 82 0 R /R112 137 0 R /F1 200 0 R /R110 141 0 R /R104 149 0 R /R74 99 0 R /R106 153 0 R /F2 201 0 R /R13 47 0 R /R15 51 0 R /R81 94 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 202 0 R 203 0 R 204 0 R 205 0 R 206 0 R 207 0 R 208 0 R 209 0 R 210 0 R 211 0 R 212 0 R 213 0 R 214 0 R 215 0 R 216 0 R 217 0 R 218 0 R 219 0 R 220 0 R 221 0 R 222 0 R 223 0 R 224 0 R 225 0 R 226 0 R 227 0 R 228 0 R 229 0 R 230 0 R ] >> endobj 9 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 231 0 R /Resources << /XObject << /R210 232 0 R /R206 234 0 R /R207 236 0 R /R204 237 0 R /R205 235 0 R /R208 239 0 R /R209 233 0 R /R203 238 0 R >> /ExtGState << /R12 36 0 R >> /Font << /R201 240 0 R /R199 244 0 R /F2 248 0 R /F1 249 0 R /R15 51 0 R /R13 47 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] /ColorSpace << /R29 27 0 R /R22 71 0 R >> >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 ] >> endobj 10 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 262 0 R /Resources << /ExtGState << /R12 36 0 R >> /Font << /F2 263 0 R /F1 264 0 R /R13 47 0 R /R15 51 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 265 0 R 266 0 R 267 0 R 268 0 R 269 0 R 270 0 R 271 0 R ] >> endobj 11 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 272 0 R /Resources << /ExtGState << /R12 36 0 R >> /Font << /F2 273 0 R /R19 58 0 R /F1 274 0 R /R13 47 0 R /R15 51 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 275 0 R 276 0 R 277 0 R 278 0 R 279 0 R 280 0 R 281 0 R 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 292 0 R 293 0 R 294 0 R 295 0 R 296 0 R 297 0 R 298 0 R 299 0 R 300 0 R 301 0 R 302 0 R 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 316 0 R 317 0 R 318 0 R 319 0 R 320 0 R 321 0 R 322 0 R 323 0 R 324 0 R 325 0 R ] >> endobj 12 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 326 0 R /Resources << /ExtGState << /R12 36 0 R >> /Font << /F2 327 0 R /R19 58 0 R /F1 328 0 R /R15 51 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 ] >> endobj 13 0 obj << /Type /Catalog /Pages 1 0 R >> endobj 14 0 obj << /Length 18612 >> stream q q q 0.10000 0 0 0.10000 0 0 cm /R12 gs 0 g q 10 0 0 10 0 0 cm BT /R13 14.34620 Tf 1 0 0 1 159.72500 675.06700 Tm [ (Compr) 17.99210 (ehension\055guided) -249.99300 (r) 18.00990 (eferring) -249.99600 (expr) 17.99460 (essions) ] TJ /R15 11.95520 Tf 28.01290 -37.85820 Td [ (Ruotian) -249.98900 (Luo) ] TJ -1.15391 -13.94690 Td (TTI\055Chicago) Tj /R17 8.96640 Tf -3.76016 -13.94800 Td (rluo\100ttic\056edu) Tj /R15 11.95520 Tf 115.39300 27.89490 Td [ (Gre) 15.00960 (gory) -250.00400 (Shakhnaro) 14.99540 (vich) ] TJ 25.88280 -13.94690 Td (TTI\055Chicago) Tj /R17 8.96640 Tf -3.76094 -13.94800 Td (greg\100ttic\056edu) Tj /R13 11.95520 Tf -174.34400 -41.04610 Td (Abstract) Tj /R19 9.96260 Tf -83.92770 -24.20900 Td [ (W) 91.98650 (e) -383 (consider) -381.98900 (g) 10.00320 (ener) 15.01960 (ation) -383.00500 (and) -382 (compr) 36.98890 (ehension) -383.01000 (of) -382.98300 (natur) 15.00730 (al) ] TJ -11.95510 -11.95510 Td [ (langua) 9.99159 (g) 10.00320 (e) -267.01900 (r) 37.01960 (eferring) -266.00600 (e) 19.99180 (xpr) 36.98160 (ession) -266.99000 (for) -266.01600 (objects) -266.99200 (in) -265.98300 (an) -267.00400 (ima) 10.01300 (g) 10.00320 (e) 15.01220 (\056) -360.01600 (Un\055) ] TJ 11.95510 TL T* [ (lik) 10.00810 (e) -305.01200 (g) 10.00320 (eneric) -305.98300 (\223ima) 9.98853 (g) 10.00320 (e) -305.01300 (captioning\224) -306.00500 (whic) 14.99870 (h) -305.00300 (lac) 20.00400 (ks) -305.00500 (natur) 15.00730 (al) -306.01500 (stan\055) ] TJ T* [ (dar) 36.99020 (d) -360.99400 (e) 15.01280 (valuation) -360.99600 (criteria\054) -389 (quality) -361.00600 (of) -360.98600 (a) -360.99400 (r) 37.01830 (eferring) -361.01100 (e) 19.99180 (xpr) 36.98160 (ession) ] TJ T* [ (may) -263.01100 (be) -263.99500 (measur) 36.99510 (ed) -263.01400 (by) -263.01400 (the) -263.98800 (r) 37.01960 (eceiver\047) 39.98110 (s) -263.00900 (ability) -262.98100 (to) -263.00300 (corr) 36.98650 (ectly) -263.99300 (infer) ] TJ 11.95630 TL T* [ (whic) 14.99870 (h) -194.98200 (object) -193.99100 (is) -195.01300 (being) -195.00800 (described\056) -290.98700 (F) 105.00600 (ollowing) -195.01200 (this) -195.00200 (intuition\054) -205.98500 (we) ] TJ 11.95470 TL T* [ (pr) 44.98390 (opose) -298.99200 (two) -298.00300 (appr) 44.99370 (oac) 14.98400 (hes) -299.00200 (to) -298.99700 (utilize) -299.01300 (models) -298.00700 (tr) 14.99010 (ained) -298.99200 (for) -298.99200 (com\055) ] TJ T* [ (pr) 36.98520 (ehension) -379.01000 (task) -378.00200 (to) -378.98300 (g) 10.00320 (ener) 15.01960 (ate) -378.98800 (better) -379.01800 (e) 19.99180 (xpr) 36.98160 (essions\056) -696.00600 (F) 45.01700 (ir) 10.01060 (st\054) -411.01500 (we) ] TJ T* [ (use) -326.99800 (a) -327 (compr) 36.98890 (ehension) -326 (module) -326.98300 (tr) 14.99140 (ained) -326.98600 (on) -326.99400 (human\055g) 9.98363 (ener) 15.01960 (ated) ] TJ T* [ (e) 19.99240 (xpr) 36.98100 (essions\054) -234.00600 (as) -229.99000 (a) -229.01600 (\223critic\224) -229.99200 (of) -229.98900 (r) 37.01960 (eferring) -230.01200 (e) 19.99180 (xpr) 36.98160 (ession) -230.01600 (g) 10.00320 (ener) 15.01960 (ator) 110.98400 (\056) ] TJ 11.95590 TL T* [ (The) -249.00200 (compr) 36.98890 (ehension) -248.99300 (module) -248.99600 (serves) -249.01200 (as) -249.00600 (a) -249.01200 (dif) 18.01660 (fer) 36.98400 (entiable) -248.99000 (pr) 44.98510 (oxy) ] TJ 11.95510 TL T* [ (of) -310.99500 (human) -310.98400 (e) 15.01280 (valuation\054) -326.01800 (pr) 44.98390 (o) 10.00320 (viding) -310.98200 (tr) 14.99140 (aining) -311.01200 (signal) -311.01200 (to) -310.99500 (the) -311 (g) 10.00320 (en\055) ] TJ T* [ (er) 15.01890 (ation) -252.98700 (module) 14.99870 (\056) -319.98300 (Second\054) -253.98300 (we) -253.00800 (use) -253.00900 (the) -253.98900 (compr) 38.00840 (ehension) -254.01200 (module) ] TJ T* [ (in) -223.99000 (a) -223.99700 (g) 10.00320 (ener) 15.01960 (ate\055and\055r) 36.99630 (er) 15.01830 (ank) -223.99600 (pipeline) 9.98730 (\054) -229.01800 (whic) 15 (h) -225.01600 (c) 15.01220 (hooses) -224.01900 (fr) 44.98510 (om) -223.99400 (can\055) ] TJ T* [ (didate) -256.99100 (e) 19.99240 (xpr) 36.98100 (essions) -256.98500 (g) 10.00320 (ener) 15.01960 (ated) -257.98400 (by) -257.01500 (a) -257.01100 (model) -257.00100 (accor) 36.98160 (ding) -256.99400 (to) -257.98400 (their) ] TJ T* [ (performance) -224.01200 (on) -225.01100 (the) -223.99400 (compr) 36.98890 (ehension) -224.99700 (task\056) -300.99100 (W) 91.98710 (e) -224.98700 (show) -224.01200 (that) -224.99700 (both) ] TJ 11.95590 TL T* [ (appr) 44.99370 (oac) 14.98340 (hes) -334.01600 (lead) -334.99100 (to) -334.01000 (impr) 44.99370 (o) 10.00320 (ved) -335.00700 (r) 37.01960 (eferring) -333.99600 (e) 19.99180 (xpr) 36.98160 (ession) -335.01800 (g) 10.00320 (ener) 15.01960 (a\055) ] TJ 11.95510 TL T* [ (tion) -250.01200 (on) -249.98800 (multiple) -250 (benc) 15.01830 (hmark) -249.98300 (datasets\056) ] TJ /R13 11.95520 Tf 36.76020 TL T* [ (1\056) -249.99000 (Intr) 18.01460 (oduction) ] TJ /R15 9.96260 Tf 11.95510 -19.22700 Td [ (Image) -366.99200 (captioning\054) -396.99100 <6465026e6564> -367.01600 (broadly) -367.01700 (as) -366.99500 (automatic) -367.98500 (gener) 19.99420 (\055) ] TJ -11.95510 -11.95470 Td [ (ation) -365.01700 (of) -366.01600 (te) 14.98100 (xt) -364.98600 (describing) -365.00900 (images\054) -393.98800 (has) -366.01000 (seen) -364.99500 (much) -365.02000 (recent) -365.98300 (at\055) ] TJ 11.95510 TL T* [ (tention\056) -658.99200 (Deep) -366.99900 (learning\054) -395.01300 (and) -366.99100 (in) -366.00500 (particular) -366.00500 (recurrent) -367.00200 (neural) ] TJ T* [ (netw) 10.00870 (orks) -261.98300 (\050RNNs\051\054) -264.01400 (ha) 19.99790 (v) 14.98280 (e) -262 (led) -261.98800 (to) -261.98300 (a) -261.02000 <7369676e690263616e74> -261.98600 (impro) 15.00480 (v) 14.98280 (ement) -262.02000 (in) ] TJ 11.95590 TL T* [ (state) -229.98900 (of) -229.01900 (the) -229.99300 (art\056) -303.00600 (Ho) 24.98360 (we) 25.01420 (v) 14.98280 (er) 39.98600 (\054) -233.99800 (the) -229.99300 (metrics) -230.01900 (currently) -229.00600 (used) -229.98900 (to) -229.98900 (e) 25.01050 (v) 24.98110 (al\055) ] TJ 11.95510 TL T* [ (uate) -327.98100 (image) -328.98800 (captioning) -327.98600 (are) -329.01600 (mostly) -327.98100 (borro) 24.98360 (wed) -327.98200 (from) -328.99400 (machine) ] TJ T* [ (translation\056) -364.99700 (This) -268.01200 (misses) -268 (the) -267.98700 (naturally) -268.98900 (multi\055m) 0.98758 (odal) -269.00200 (distrib) 19.99180 (u\055) ] TJ T* [ (tion) -250.01200 (of) -249.99600 (appropriate) -249.98300 (captions) -250.01000 (for) -249.99900 (man) 14.99010 (y) -249.99300 (scenes\056) ] TJ 11.95510 -12.25270 Td [ (Referring) -358.98700 (e) 15.01220 (xpressions) -357.98100 (are) -359.01100 (a) -359.00400 (special) -358.98700 (case) -357.98700 (of) -358.99600 (image) -358.98400 (cap\055) ] TJ -11.95510 -11.95630 Td [ (tions\056) -436.99200 (Such) -292.00900 (e) 15.01220 (xpressions) -291.99200 (describe) -291.98300 (an) -293.00900 (object) -292.01500 (or) -291.98800 (re) 15.00980 (gion) -292.00800 (in) -292.99800 (the) ] TJ T* [ (image\054) -268.99600 (with) -266.01200 (the) -265.00700 (goal) -265.98300 (of) -265.01300 (identifying) -265.98700 (it) -265 (uniquely) -266 (to) -265.00200 (a) -266 (listener) 54.99080 (\056) ] TJ T* [ (Thus\054) -228.01800 (in) -221.99000 (contrast) -221.98400 (to) -221.99000 (generic) -222.98100 (captioning\054) -226.98000 (referring) -222.98700 (e) 15.01220 (xpression) ] TJ T* [ (generation) -260.99700 (has) -259.98800 (a) -261.02000 (natural) -261.00300 (e) 25.01050 (v) 24.98110 (aluation) -260.98700 (metric\072) -330.98200 (a) -261.02000 (human) -260.99300 (should) ] TJ T* [ (easily) -213.98700 (comprehend) -213.00200 (the) -213.99600 (description) -214.00700 (and) -213.01700 (identify) -214.01300 (the) -213.99600 (object\050s\051) ] TJ 11.95590 TL T* [ (being) -250.01900 (described\056) ] TJ 11.95510 -12.25310 Td [ (In) -305.00600 (this) -304.98200 (paper) 39.99090 (\054) -319.00300 (we) -305.00900 (consider) -304.99200 (tw) 10.00810 (o) -305.00300 (related) -305.00500 (tasks\056) -476.01600 (One) -305.00500 (is) -304.99600 (the) ] TJ -11.95510 -11.95510 Td [ (generation) -383.99600 (task\072) -576.98100 (generating) -383.99600 (a) -384.01900 (discriminati) 24.98850 (v) 14.98280 (e) -384.01900 (referring) -384.01900 (e) 15.01220 (x\055) ] TJ 11.95510 TL T* [ (pression) -335.01500 (for) -335.00400 (an) -333.98200 (object) -334.98700 (in) -334.99100 (an) -335.00200 (image\056) -564.00500 (The) -334.99800 (other) -334.99300 (is) -334.99100 (the) ] TJ /R19 9.96260 Tf 216.33500 0 Td (com\055) Tj ET Q q 3279.88000 4215.97000 1979.95000 1466.71000 re W n /R22 cs 1 1 1 scn 2853.37000 3378.16000 4373.91000 2460.33000 re f 0 0 0 scn q 10 0 0 10 0 0 cm BT 0.25247 Tc /R23 7.40376 Tf 1 0 0 1 424.93800 556.72300 Tm [ (Ma) 16.98220 (x) -4.00319 ( ) 129.98800 (l) 3.97681 (i) 3.97681 (k) -4.00319 (e) 16.97560 (l) 3.97681 (i) 3.97681 (h) -4.00319 (o) 72.97420 (o) -4.00319 (d) 72.97420 (\072) ] TJ -0.12957 Tc /R25 7.40376 Tf 50.11760 0 Td [ (bo) -55.99850 (w) 12.02280 (l) -46.98970 ( ) 79.02180 (o) -56.02490 (f) 8.01298 ( ) 79.02180 (f) 8.01298 (o) -56.02490 (o) -56.02490 (d ) 78.99540 (o) -56.02490 (n ) ] TJ 0.21989 Tc -50.11760 -9.11250 Td [ (le) 11.99640 (f) 55.00270 (t) ] TJ 0.25247 Tc /R23 7.40376 Tf 8.54258 TL T* [ (MMI) 58.98610 (\072) ] TJ -0.12957 Tc /R25 7.40376 Tf 19.36370 0 Td [ (bo) -55.99850 (w) 12.02280 (l) -46.98970 ( ) 1.99170 (o) -56.02490 (f) 8.01298 ( ) 1.99170 (f) 8.01298 (o) -56.02490 (o) -56.02490 (d ) 78.99540 (o) -56.02490 (n ) 2.01808 (l) -46.98970 (e) -34.99330 (f) 8.01298 (t) ] TJ /R23 7.40376 Tf -19.36370 -17.65510 Td [ (Pr) 7.98660 (o) -56.02490 (x) -56.02490 (y) -56.02490 (\072) ] TJ /R25 7.40376 Tf 21.64180 0 Td [ (bo) -55.99850 (w) 12.02280 (l) -46.98970 ( ) 79.02180 (o) -56.02490 (f) 8.01298 ( ) 1.99170 (f) 8.01298 (o) -56.02490 (o) -56.02490 (d ) 2.01808 (o) -56.02490 (n ) 2.01808 (l) -46.98970 (e) -34.99330 (f) 8.01298 (t) 8.01298 ( ) ] TJ -21.64180 -9.11250 Td [ (bo) -55.99850 (t) 8.01298 (t) 8.01298 (o) -56.02490 (m) -30.99010 ( ) 1.99170 (l) -46.98970 (e) -34.99330 (f) 8.01298 (t) ] TJ 0.18732 Tc /R23 7.40376 Tf T* [ (Re) 7.99319 (r) 51.02590 (a) 8.01298 (n) -13.01200 (k) ] TJ 0 Tc 22.21130 0 Td (\072) Tj -0.12957 Tc /R25 7.40376 Tf 3.41680 0 Td [ (bo) -55.99850 (t) 8.01298 (t) 8.01298 (o) -56.02490 (m) -30.99010 ( ) 79.02180 (l) -46.98970 (e) -34.99330 (f) 8.01298 (t) 8.01298 ( ) 1.99170 (bo) -55.99850 (w) 12.02280 (l) ] TJ ET Q Q q 3320.38000 5035.46000 877.06300 580.91000 re W n Q q 3320.38000 5035.46000 877.06300 580.91000 re W n q 1041.90000 0 0 724.57100 3248.29000 4950.37000 cm /R31 Do Q Q q 3279.88000 4215.97000 1979.95000 1466.71000 re W n /R22 cs 0 0 0 scn q 10 0 0 10 0 0 cm BT 0.25247 Tc /R23 7.40376 Tf 1 0 0 1 424.93800 477.42900 Tm [ (Ma) 16.98220 (x) -4.00319 ( ) 129.98800 (l) 3.97681 (i) 3.97681 (k) -4.00319 (e) 16.97560 (l) 3.97681 (i) 3.97681 (h) -4.00319 (o) 72.97420 (o) -4.00319 (d) 72.97420 (\072) ] TJ -0.12957 Tc /R25 7.40376 Tf 50.11800 0 Td [ (bi) -47.00950 (r) -34.99330 (d ) 78.99540 (w) 12.02280 (i) -46.98970 (t) 8.01298 (h ) 2.01808 (g) -56.02490 (r) -34.99330 (e) -34.99330 (e) -34.99330 (n) ] TJ 0.25247 Tc /R23 7.40376 Tf -50.11800 -9.11211 Td [ (MMI) 58.98610 (\072) 3.97681 ( ) ] TJ -0.12957 Tc /R25 7.40376 Tf 19.36370 0 Td [ (bi) -47.00950 (r) -34.99330 (d ) 2.01808 (w) 12.02280 (i) -46.98970 (t) 8.01298 (h ) -75.01200 (g) -56.02490 (r) -34.99330 (e) -34.99330 (e) -34.99330 (n ) 78.99540 (l) -46.98970 (e) -34.99330 (a) -56.02490 (v) -56.02490 (e) -34.99330 (s) ] TJ /R23 7.40376 Tf -19.36370 -17.65510 Td [ (Pr) 7.98660 (o) -56.02490 (x) -56.02490 (y) -56.02490 (\072) ] TJ /R25 7.40376 Tf 21.64180 0 Td [ (bl) -47.00950 (ur) -35.01970 (r) -34.99330 (y) -56.02490 ( ) 79.02180 (bi) -47.00950 (r) -34.99330 (d) ] TJ 0.18732 Tc /R23 7.40376 Tf -21.64180 -8.54297 Td [ (Re) 7.99319 (r) 51.02590 (a) 8.01298 (n) -13.01200 (k) ] TJ 0 Tc 22.21130 0 Td (\072) Tj -0.12957 Tc /R25 7.40376 Tf 3.41719 0 Td [ (bl) -47.00950 (ur) -35.01970 (r) -34.99330 (y) -56.02490 ( ) 79.02180 (bi) -47.00950 (r) -34.99330 (d) ] TJ ET Q Q q 3320.38000 4272.31000 877.06300 700.50800 re W n Q q 3320.38000 4272.31000 877.06300 700.50800 re W n q 993.67900 0 0 1003.72000 3303 4018.05000 cm /R33 Do Q Q q 10 0 0 10 0 0 cm BT /R15 9.96260 Tf 1 0 0 1 308.86200 403.26600 Tm [ (Figure) -270.00900 (1\072) -350.00800 (F) 14.99260 (or) -269.99200 (each) -270.00200 (image\054) -274.99300 (the) -269.98400 (top) -270.99400 (tw) 10.00810 (o) -269.98900 (e) 15.01220 (xpressions) -269.99400 (are) -270.00400 (gen\055) ] TJ 11.95510 TL T* [ (erated) -240.98400 (by) -241.00900 (baseline) -241.00100 (models) -240.00900 (proposed) -240.99400 (in) -241.00900 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R15 9.96260 Tf 1 0 0 1 466.58000 391.31100 Tm (23) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R15 9.96260 Tf 1 0 0 1 476.54300 391.31100 Tm [ (\135\073) -243.99400 (the) -241.00900 (bottom) -241.01900 (tw) 10.00810 (o) ] TJ -167.68000 -11.95590 Td [ (e) 15.01220 (xpressions) -250 (are) -250.01000 (generated) -249.99700 (by) -249.98800 (our) -249.99300 (methods\056) ] TJ /R19 9.96260 Tf 34.40820 TL T* [ (pr) 36.98650 (ehension) ] TJ /R15 9.96260 Tf 46.87110 0 Td [ (task) -298.01400 (\050called) -297.00700 (natural) -298.01400 (language) -298.01400 (object) -296.99500 (retrie) 25.01540 (v) 24.98110 (al) -297.98500 (in) ] TJ -46.87110 -11.95470 Td (\133) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R15 9.96260 Tf 1 0 0 1 312.18000 332.99200 Tm (15) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R15 9.96260 Tf 1 0 0 1 322.14200 332.99200 Tm [ (\135\051\072) -309.98700 (localizing) -248.98500 (an) -249.01700 (object) -249.00300 (in) -249.98800 (an) -249.01700 (image) -249.00300 (gi) 24.98600 (v) 14.98280 (en) -249.01700 (a) -248.98300 (referring) -250.00200 (e) 15.01220 (x\055) ] TJ -13.28010 -11.95510 Td [ (pression\056) -309.99700 (Most) -248.00800 (prior) -249.00700 (w) 10 (ork) -249.01200 (addressed) -247.98800 (both) -248.99300 (tasks) -248.99800 (by) -249.00700 (b) 20.00160 (uilding) ] TJ 11.95510 TL T* [ (a) -359.98400 (sequence) -360.99400 (generation) -360.00100 (model\056) -641.98100 (Such) -359.99900 (a) -359.98400 (model) -361.01300 (can) -359.98400 (be) -360.99900 (used) ] TJ T* [ (discriminati) 24.98850 (v) 14.98280 (ely) -392.98600 (for) -394.01500 (the) -392.98600 (comprehension) -392.99300 (task\054) -429.01200 (by) -394.00300 (inferring) ] TJ 11.95590 TL T* [ (the) -249.99000 (re) 15.00980 (gion) -250.01500 (which) -249.98300 (maximizes) -249.99700 (the) -249.98800 (e) 15.01220 (xpression) -250.00700 (posterior) 54.99570 (\056) ] TJ 11.95510 -12.42190 Td [ (In) -503.01200 (contrast\054) -564.98200 (we) -503.01700 (dra) 15.01470 (w) -503.00900 (inspiration) -501.98500 (from) -503.00400 (the) -503.00400 (generator) 20.00160 (\055) ] TJ -11.95510 -11.95510 Td [ (discriminator) -1277 (structure) -1276.01000 (in) -1276.99000 (Generati) 24.98110 (v) 14.98280 (e) -1277.01000 (Adv) 14.99260 (ersarial) ] TJ 11.95510 TL T* [ (Netw) 10.00570 (orks\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R15 9.96260 Tf 1 0 0 1 350.81400 248.83900 Tm (10) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R15 9.96260 Tf 1 0 0 1 360.77700 248.83900 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R15 9.96260 Tf 1 0 0 1 369.95200 248.83900 Tm (26) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R15 9.96260 Tf 1 0 0 1 379.91500 248.83900 Tm [ (\135\056) -1572.99000 (In) -670.98600 (GANs\054) -776.99300 (the) -671.01600 (generator) -670.98100 (module) ] TJ -71.05270 -11.95510 Td [ (tries) -485.98500 (to) -487.00500 (generate) -486 (a) -486.98200 (signal) -486.01200 (\050e\056g\056\054) -545.98200 (natural) -487.00200 (image\051\054) -544.99200 (and) -487.01200 (the) ] TJ 11.95510 TL T* [ (discriminator) -416.00200 (module) -416.01900 (tries) -415.99400 (to) -416 (tell) -415.98900 (real) -416.01400 (images) -415.98900 (apart) -416.00900 (from) ] TJ T* [ (the) -374.98900 (generated) -376.01300 (ones\056) -685.98800 (F) 14.99260 (or) -376.01300 (our) -374.99100 (task\054) -407.01500 (the) -374.98600 (generator) -376.01100 (produces) ] TJ 11.95630 TL T* [ (referring) -470.00500 (e) 15.01220 (xpressions\056) -969.99500 (W) 79.98660 (e) -470.00500 (w) 10.00320 (ould) -470.99500 (lik) 10.01790 (e) -470.00500 (these) -469.99000 (e) 15.01220 (xpressions) ] TJ 11.95470 TL T* [ (to) -392 (be) -390.99400 (both) -391.99100 <696e74656c6c696769626c652f0375656e74> -390.97900 (and) -392.00800 (unambiguous) -391.01800 (to) -392.00300 (human\056) ] TJ T* [ (Fluenc) 15 (y) -467.01500 (can) -466.00300 (be) -467.02000 (encouraged) -467.01500 (by) -467.01000 (using) -466.01000 (the) -467.01000 (standard) -467.00500 (cross) ] TJ T* [ (entrop) 10.01300 (y) -257.99100 (loss) -258.01100 (with) -258.01300 (respect) -257.99400 (to) -259.00600 (human\055generated) -258.01600 (e) 15.01220 (xpressions\051\056) ] TJ T* [ (On) -412.00200 (the) -410.98300 (other) -412 (hand\054) -451.99100 (we) -412.01200 (adopt) -411.99000 (a) -410.99500 (comprehension) -412.01000 (model) -411.98500 (as) ] TJ T* [ (the) -467.01300 (\223discriminator\224) -466 (which) -467.00500 (tells) -466.01500 (if) -467.01500 (the) -467.01000 (e) 15.01220 (xpres) 0.99003 (sion) -466.99600 (can) -466.98600 (be) ] TJ 11.95590 TL T* [ (correctly) -382.01000 (dereferenced\056) -706.99700 (Note) -382.00500 (that) -382 (we) -382.01500 (can) -382.02000 (also) -382.98000 (re) 15.00730 (g) 4.98446 (ard) -382.02000 (the) ] TJ 11.95510 TL T* [ (comprehension) -259.01600 (model) -258.99100 (as) -259.01300 (a) -258.98100 (\223critic\224) -259.01600 (of) -259.01600 (the) -259.00600 (\223action\224) -259.01100 (made) -259.00600 (by) ] TJ T* [ (the) -249.99000 (generator) -249.99500 (where) -250.00200 (the) -249.99000 (\223action\224) -249.99300 (is) -249.98300 (each) -250.00700 (generated) -249.99700 (w) 10.00320 (ord\056) ] TJ 11.95510 -12.42190 Td [ (Instead) -386.98200 (of) -388.01100 (an) -386.99400 (adv) 14.98280 (ersarial) -386.99400 (relationship) -386.98900 (between) -386.98900 (the) -388.00400 (tw) 10.00810 (o) ] TJ -11.95510 -11.95510 Td [ (modules) -386.01600 (in) -384.98200 (GANs\054) -419.99100 (our) -384.98700 (architecture) -386.00900 (is) -386 (c) 1.01454 (ollaborati) 25.00560 (v) 14.98280 (e) -386.01900 (\226) -386.00900 (the) ] 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 (7102) 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 /FlateDecode
/Interpolate true
/Length 636
/ColorSpace /DeviceGray
/DecodeParms <<
/Columns 322
/Predictor 15
>>
/BitsPerComponent 8
/Height 326
/Width 322
/Subtype /Image
>>
stream
x1 ZCMV*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YBV!U*d
YF
endstream
endobj
19 0 obj
<<
/Filter /FlateDecode
/Resources <<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
/XObject <<
/x15 20 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
20 0 obj
<<
/Filter /FlateDecode
/Resources 21 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
21 0 obj
<<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
>>
endobj
22 0 obj
<<
/Filter /FlateDecode
/Interpolate true
/Length 526
/ColorSpace /DeviceGray
/DecodeParms <<
/Columns 356
/Predictor 15
>>
/BitsPerComponent 8
/Height 250
/Width 356
/Subtype /Image
>>
stream
xA 0 ?L F=Ɂ
endstream
endobj
23 0 obj
<<
/Filter /FlateDecode
/Resources <<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
/XObject <<
/x24 24 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
24 0 obj
<<
/Filter /FlateDecode
/Resources 25 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