%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 (Speaking the Same Language\072 Matching Machine to Human Captions by Adversarial Training) /Producer (PyPDF2) /Author (Rakshith Shetty\054 Marcus Rohrbach\054 Lisa Anne Hendricks\054 Mario Fritz\054 Bernt Schiele) /Subject (2017 IEEE International Conference on Computer Vision) >> endobj 3 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 14 0 R /Resources << /XObject << /x8 15 0 R /R25 18 0 R /x6 19 0 R /R28 22 0 R /R27 23 0 R /x12 24 0 R /x10 27 0 R /R26 30 0 R >> /ExtGState << /s9 31 0 R /s11 34 0 R /a0 << /CA 1 /ca 1 >> /R14 37 0 R /s5 38 0 R /s7 41 0 R >> /Font << /F2 44 0 R /F1 45 0 R /R15 46 0 R /R17 49 0 R /R21 53 0 R /R19 57 0 R /R23 60 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 64 0 R 65 0 R 66 0 R 67 0 R 68 0 R ] >> endobj 4 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 69 0 R /Resources << /ExtGState << /R14 37 0 R >> /Font << /F2 70 0 R /R66 71 0 R /R15 46 0 R /R17 49 0 R /F1 74 0 R /R70 75 0 R /R23 60 0 R /R68 79 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 83 0 R 84 0 R 85 0 R 86 0 R 87 0 R 88 0 R 89 0 R 90 0 R 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 113 0 R ] >> endobj 5 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 114 0 R /Resources << /XObject << /R113 115 0 R >> /ExtGState << /R112 116 0 R /R110 117 0 R /R111 118 0 R /R14 37 0 R >> /Font << /R108 119 0 R /R70 75 0 R /R102 124 0 R /R104 127 0 R /R106 130 0 R /F2 134 0 R /F1 135 0 R /R15 46 0 R /R17 49 0 R /R23 60 0 R /R68 79 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 ] >> endobj 6 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 159 0 R /Resources << /XObject << /R113 115 0 R >> /ExtGState << /R110 117 0 R /R14 37 0 R >> /Font << /R108 119 0 R /R141 160 0 R /R143 163 0 R /R70 75 0 R /R102 124 0 R /R104 127 0 R /R106 130 0 R /F2 167 0 R /F1 168 0 R /R15 46 0 R /R17 49 0 R /R139 169 0 R /R68 79 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 ] >> endobj 7 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 191 0 R /Resources << /ExtGState << /R14 37 0 R >> /Font << /R108 119 0 R /R143 163 0 R /R70 75 0 R /R102 124 0 R /R104 127 0 R /R106 130 0 R /F2 192 0 R /F1 193 0 R /R15 46 0 R /R17 49 0 R /R139 169 0 R /R23 60 0 R /R68 79 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 199 0 R 200 0 R 201 0 R 202 0 R 203 0 R 204 0 R 205 0 R ] >> endobj 8 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 206 0 R /Resources << /ExtGState << /R14 37 0 R >> /Font << /F2 207 0 R /R108 119 0 R /F1 208 0 R /R15 46 0 R /R17 49 0 R /R70 75 0 R /R23 60 0 R /R68 79 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 ] >> endobj 9 0 obj << /Parent 1 0 R /Rotate 0 /Contents 220 0 R /Resources << /XObject << /R229 221 0 R /R224 255 0 R /R234 259 0 R /R233 260 0 R /R232 261 0 R /R226 262 0 R /R231 265 0 R /R221 268 0 R /R218 273 0 R >> /ExtGState << /R112 116 0 R /R212 222 0 R /R14 37 0 R /R187 223 0 R >> /Font << /R108 119 0 R /R190 274 0 R /R219 269 0 R /R222 256 0 R /R102 124 0 R /R225 263 0 R /R230 266 0 R /F2 276 0 R /F1 277 0 R /R15 46 0 R /R17 49 0 R /R227 224 0 R /R68 79 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /Group 254 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 278 0 R 279 0 R 280 0 R ] /Type /Page >> endobj 10 0 obj << /Parent 1 0 R /Rotate 0 /Contents 281 0 R /Resources << /XObject << /R258 282 0 R >> /ExtGState << /R14 37 0 R /R187 223 0 R >> /Font << /R250 283 0 R /F2 292 0 R /F1 293 0 R /R15 46 0 R /R17 49 0 R /R23 60 0 R /R70 75 0 R /R68 79 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /Group 254 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 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 ] /Type /Page >> endobj 11 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 304 0 R /Resources << /ExtGState << /R14 37 0 R >> /Font << /R66 71 0 R /F2 305 0 R /F1 306 0 R /R15 46 0 R /R17 49 0 R /R23 60 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 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 ] >> endobj 12 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 352 0 R /Resources << /ExtGState << /R14 37 0 R >> /Font << /F2 353 0 R /R23 60 0 R /F1 354 0 R /R17 49 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 367 0 R 368 0 R 369 0 R 370 0 R 371 0 R 372 0 R 373 0 R 374 0 R 375 0 R 376 0 R 377 0 R 378 0 R ] >> endobj 13 0 obj << /Type /Catalog /Pages 1 0 R >> endobj 14 0 obj << /Length 16219 >> stream q q q 0.10000 0 0 0.10000 0 0 cm /R14 gs 0 g q 10 0 0 10 0 0 cm BT /R15 14.34620 Tf 1 0 0 1 204.95100 675.06700 Tm [ (Speaking) -249.98800 (the) -250.00800 (Same) -250.00200 (Language\072) ] TJ -103.01200 -17.93320 Td [ (Matching) -249.99200 (Machine) -250.00700 (to) -250.01200 (Human) -249.99000 (Captions) -249.99600 (by) -250.01300 (Adv) 10.00140 (ersarial) -250.01200 (T) 73.98990 (raining) ] TJ /R17 11.95520 Tf 38.85000 -37.85780 Td [ (Rakshith) -250.00600 (Shetty) ] TJ /R19 7.97010 Tf 76.05700 4.33867 Td [ (1) -0.30019 ] TJ /R17 11.95520 Tf 19.67700 -4.33867 Td [ (Marcus) -250.01200 (Rohrbach) ] TJ /R19 7.97010 Tf 85.32300 4.33867 Td [ (2) -0.30019 ] TJ /R21 7.97010 Tf 4.23398 0 Td [ (\073) -0.10109 ] TJ /R19 7.97010 Tf 2.35312 0 Td [ (3) -0.30019 ] TJ /R17 11.95520 Tf 19.67580 -4.33867 Td [ (Lisa) -250.01000 (Anne) -249.98500 (Hendricks) ] TJ /R19 7.97010 Tf 101.59400 4.33867 Td [ (2) -0.30019 ] TJ /R17 11.95520 Tf -224.06400 -23.95590 Td [ (Mario) -250.01000 (Fritz) ] TJ /R19 7.97010 Tf 54.78910 4.33906 Td [ (1) -0.30019 ] TJ /R17 11.95520 Tf 19.67700 -4.33906 Td [ (Bernt) -249.98900 (Schiele) ] TJ /R19 7.97010 Tf 64.74800 4.33906 Td [ (1) -0.30019 ] TJ -288.92300 -25.28590 Td [ (1) -0.30019 ] TJ /R17 11.95520 Tf 4.73281 -4.33906 Td [ (Max) -250.00500 (Planck) -250 (Institute) -250.01500 (for) -249.99600 (Informatics\054) -249.99500 (Saarland) -249.98900 (Informatics) -250.00600 (Campus\054) -250.00400 (Saarbr) ] TJ 360.91600 0.06016 Td (\250) Tj -0.99805 -0.06016 Td [ (uck) 10.00230 (en\054) -249.99700 (German) 14.98920 (y) ] TJ /R19 7.97010 Tf -315.14400 -12.44410 Td [ (2) -0.29866 ] TJ /R17 11.95520 Tf 4.73203 -4.33789 Td [ (UC) -249.99600 (Berk) 9.98907 (ele) 14.99020 (y) -250 (EECS\054) -250.01200 (CA\054) -249.98500 (United) -250 (States) ] TJ /R19 7.97010 Tf 226.17800 4.33789 Td [ (3) -0.30019 ] TJ /R17 11.95520 Tf 4.73320 -4.33789 Td [ (F) 15.01580 (acebook) -249.99300 (AI) -250.00200 (Research) ] TJ /R15 11.95520 Tf -215.08500 -43.88090 Td (Abstract) Tj /R23 9.96260 Tf -83.92810 -24.95430 Td [ (While) -284.99500 (str) 44.99250 (ong) -286.01600 (pr) 44.98390 (o) 10.00320 (gr) 36.98520 (ess) -285.00400 (has) -286.01500 (been) -285.01000 (made) -285.98400 (in) -285 (ima) 10.01300 (g) 10.00320 (e) -285.99600 (caption\055) ] TJ -11.95470 -11.95470 Td [ (ing) -195.98900 (r) 37.01960 (ecently) 55.01230 (\054) -206.98300 (mac) 14.98160 (hine) -195.99400 (and) -195.99200 (human) -195.98300 (captions) -196.00900 (ar) 36.98650 (e) -196.99200 (stil) 1.00718 (l) -197.01900 (quite) -195.98700 (dis\055) ] TJ 11.95510 TL T* [ (tinct\056) -377.01100 (This) -272.98100 (is) -271.98000 (primarily) -271.99100 (due) -273.00700 (to) -271.98200 (the) -273.00600 <6465026369656e63696573> -272.01800 (in) -271.98200 (the) -273.00600 (g) 10.00320 (ener) 20 (\055) ] TJ T* [ (ated) -296.99600 (wor) 36.99870 (d) -297.00400 (distrib) 20.00160 (ution\054) -309.01500 (vocab) 20.00280 (ulary) -296.99100 (size) 10.01300 (\054) -309.00500 (and) -296.99500 (str) 44.99250 (ong) -296.99500 (bias) -296.98700 (in) ] TJ 11.95590 TL T* [ (the) -213.01500 (g) 10.00320 (ener) 15.01960 (ator) 9.98240 (s) -213.01700 (towar) 37.01100 (ds) -213.01200 (fr) 36.98770 (equent) -213.99000 (capt) 0.99248 (ions\056) -298.00200 (Furthermor) 37.01830 (e) 9.99343 (\054) -221.01900 (hu\055) ] TJ 11.95510 TL T* [ (mans) -308.00800 (\226) -309.00200 (rightfully) -308.01400 (so) -308.99600 (\226) -307.98300 (g) 10.00320 (ener) 15.01960 (ate) -308.02000 (multiple) 9.99588 (\054) -323.00200 (diver) 10.00810 (se) -309.00500 (captions\054) ] TJ T* [ (due) -301.00300 (to) -300.99600 (the) -301.98100 (inher) 36.99260 (ent) -301.00100 (ambiguity) -301.00700 (in) -300.99600 (the) -301.00100 (captioning) -301.98100 (tas) 0.99738 (k) -301.99400 (whic) 15 (h) ] TJ T* [ (is) -249.98400 (not) -250.02000 (e) 19.99240 (xplicitly) -250.00900 (consider) 37 (ed) -249.99600 (in) -249.98500 (today\047) 40.00800 (s) -249.99100 (systems\056) ] TJ 11.95470 -12.99880 Td [ (T) 91.99870 (o) -271.98900 (addr) 36.99510 (ess) -271.98600 (these) -271.98400 (c) 15.01220 (halleng) 9.98975 (es\054) -277.98900 (we) -271.98500 (c) 15.01220 (hang) 10.01790 (e) -271.99900 (the) -271.98600 (tr) 14.99010 (aining) -271.99900 (ob\055) ] TJ -11.95470 -11.95510 Td [ (jective) -329 (of) -328.99200 (the) -328.01600 (caption) -329.01300 (g) 10.00320 (ener) 15.01960 (ator) -328.98100 (fr) 44.98640 (om) -328.99700 (r) 37.01830 (epr) 45.01940 (oducing) -329.01600 (gr) 44.98510 (ound\055) ] TJ T* [ (truth) -418.01700 (captions) -418.01200 (to) -417.99600 (g) 10.00320 (ener) 15.01960 (ating) -418.99900 (a) -418.00400 (set) -418 (of) -417.99600 (captions) -418.01100 (that) -419.00400 (is) -417.99600 (in\055) ] TJ T* [ (distinguishable) -410.01000 (fr) 44.98640 (om) -410.00300 (human) -409.98700 (written) -410.00800 (captions\056) -791.00500 (Instead) -409.99000 (of) ] TJ 11.95630 TL T* [ (handcr) 15 (afting) -339.98500 (suc) 14.98520 (h) -340.99700 (a) -340.01700 (learning) -341.00600 (tar) 36.99260 (g) 10.00320 (et\054) -362.99300 (we) -340.01200 (employ) -340.99000 (adver) 10.00570 (sar) 20.01380 (\055) ] TJ 11.95470 TL T* [ (ial) -395.99300 (tr) 14.99140 (aining) -396.01700 (in) -396 (combination) -396.00100 (with) -396.01800 (an) -396.98300 (appr) 44.99490 (oximate) -395.99300 (Gumbel) ] TJ T* [ (sampler) -208.01600 (to) -207.99200 (implicitly) -206.98300 (matc) 14.98890 (h) -208 (the) -207.99700 (g) 10.00320 (ener) 15.01960 (ated) -207.99200 (distrib) 20.00160 (ution) -208.01400 (to) -207.99200 (the) ] TJ T* [ (human) -226.01800 (one) 14.98340 (\056) -301.98600 (While) -225.98400 (our) -225.98600 (method) -225.98100 (ac) 15.01830 (hie) 14.98520 (ves) -226.98400 (compar) 14.99750 (able) -225.98900 (perfor) 19.98200 (\055) ] TJ T* [ (mance) -299.98500 (to) -300.01600 (the) -299 (state\055of\055the\055art) -300.01700 (in) -300.01600 (terms) -298.99100 (of) -300.01600 (the) -299.98200 (corr) 36.98650 (ectness) -300.01600 (of) ] TJ T* [ (the) -275.00500 (captions\054) -281.99300 (we) -275.00400 (g) 10.00320 (ener) 15.01960 (ate) -275.98600 (a) -275.00800 (set) -275.00400 (of) -275.98100 (diver) 10.00810 (se) -275.01000 (captions) -275.01500 (that) -276.00800 (ar) 36.98650 (e) ] TJ 11.95590 TL T* [ <7369676e690263616e746c79> -352.00800 (less) -350.98600 (biased) -351.99500 (and) -350.98600 (better) -352.00300 (matc) 14.98770 (h) -352.01500 (t) 0.98758 (h) -1.01454 (e) -351.00500 (global) -351.98500 (uni\055\054) ] TJ 11.95510 TL T* [ (bi\055) -249.98800 (and) -249.98200 (tri\055gr) 15.00610 (am) -249.99000 (distrib) 20.00160 (utions) -250.00100 (of) -249.98500 (the) -249.99000 (human) -250.01500 (captions\056) ] TJ /R15 11.95520 Tf 38.99570 TL T* [ (1\056) -249.99000 (Intr) 18.01460 (oduction) ] TJ /R17 9.96260 Tf 11.95470 -19.97300 Td [ (Image) -273.00800 (captioning) -273.01400 (systems) -273.01400 (ha) 19.99790 (v) 14.98280 (e) -273.01800 (a) -273.01800 (v) 24.98110 (ariety) -273.00600 (of) -273.01100 (applications) ] TJ -11.95470 -11.95510 Td [ (ranging) -324.98500 (from) -324.99400 (media) -326.00800 (retrie) 25.01420 (v) 24.98110 (al) -325.00200 (and) -324.99800 (tagging) -325.01400 (to) -326.01200 (ass) 0.98758 (istance) -326.01200 (for) ] TJ 11.95510 TL T* [ (the) -255.98900 (visually) -255.99700 (impaired\056) -329.01600 (In) -255.99500 (particular) 40.00560 (\054) -257.99400 (models) -256.00400 (which) -257.00100 (combine) ] TJ T* [ (state\055of\055the\055art) -231.98900 (image) -231.98500 (representations) -232.01000 (based) -231.99300 (on) -231.99100 (deep) -231.99800 (con) 39.99820 (v) 20.00160 (o\055) ] TJ T* [ (lutional) -243.99900 (netw) 10.00870 (orks) -243.98600 (and) -243.99200 (deep) -243.99700 (recurrent) -244.00500 (language) -245.00300 (models) -244.00600 (ha) 19.99670 (v) 14.98280 (e) ] TJ 11.95590 TL T* [ (led) -400.98400 (to) -402 (e) 25.01110 (v) 14.98280 (er) -401 (increasing) -401.99200 (performance) -400.99200 (on) -400.98200 (e) 25.01050 (v) 24.98110 (aluation) -401.98400 (metrics) ] TJ 11.95510 TL T* [ (such) -328.01200 (as) -327.98300 (CIDEr) -328.01000 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R17 9.96260 Tf 1 0 0 1 116.35400 129.86400 Tm (39) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R17 9.96260 Tf 1 0 0 1 126.31600 129.86400 Tm [ (\135) -327.98800 (and) -328.01800 (METEOR) -327.98400 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R17 9.96260 Tf 1 0 0 1 198.09600 129.86400 Tm (8) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R17 9.96260 Tf 1 0 0 1 203.07800 129.86400 Tm [ (\135) -327.98700 (as) -327.98200 (can) -329.00600 (be) -327.98400 (seen) -327.98200 (e\056g\056) ] TJ -152.96600 -11.95470 Td [ (on) -249.98800 (the) -249.99000 (COCO) -250.00500 (image) -249.98200 (Caption) -249.99900 (challenge) -249.98500 (leaderboard) -249.99500 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R17 9.96260 Tf 1 0 0 1 261.77700 117.90900 Tm (6) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R17 9.96260 Tf 1 0 0 1 266.75800 117.90900 Tm (\135\056) Tj -204.69100 -12.99920 Td [ (Despite) -375.01600 (these) -374.00600 (adv) 24.98240 (ances\054) -406.01100 (it) -374.98200 (is) -374.98300 (often) -374.98600 (easy) -374.01400 (for) -374.99900 (humans) -375.00600 (to) ] TJ -11.95470 -11.95510 Td [ (dif) 24.98540 (ferentiate) -216.98100 (between) -217.99500 (machine) -217.98700 (and) -217.01600 (human) -218.02000 (captions) -216.99600 (\226) -217.99800 (partic\055) ] TJ 11.95510 TL T* [ (ularly) -239.00200 (when) -239.01000 (observing) -238.98800 (multi) 1.01944 (p) -1.01454 (l) 0.98758 (e) -238.98500 (captions) -238.99200 (for) -238.98200 (a) -238.98500 (single) -238.99200 (image\056) ] TJ ET Q q 944.89600 0 0 590.58700 3189.62000 4646.10000 cm /R25 Do Q q 945.03000 0 0 590.62900 4329.65000 4646.10000 cm /R26 Do Q q 10 0 0 10 0 0 cm BT /R15 7.97010 Tf 1 0 0 1 318.96200 455.14600 Tm (Ours) Tj /R17 7.97010 Tf 17.27110 0 Td [ (\072) -326.01400 (a) -257.98900 (person) -258.99400 (on) -258.02300 (skis) -258.00800 (jumping) ] TJ -17.27110 -9.46523 Td [ (o) 14.98520 (v) 14.98520 (er) -249.98200 (a) -250 (ramp) ] TJ /R15 7.97010 Tf 114.00300 9.46523 Td (Ours) Tj /R17 7.97010 Tf 17.27110 0 Td [ (\072) -507.01600 (a) -348.02300 (skier) -348.99100 (is) -348.02300 (making) -348.01100 (a) -349.00300 (turn) ] TJ -17.27110 -9.46523 Td [ (on) -249.98200 (a) -250 (course) ] TJ ET Q q 944.93800 0 0 590.65600 3189.62000 3837.80000 cm /R27 Do Q q 944.89600 0 0 590.58000 4329.65000 3837.80000 cm /R28 Do Q q 10 0 0 10 0 0 cm BT /R15 7.97010 Tf 1 0 0 1 318.96200 374.31500 Tm (Ours) Tj /R17 7.97010 Tf 17.27110 0 Td [ (\072) -1030.01000 (a) -611.01800 (cros) 0.98635 (s) -610.97800 (country) -610.01600 (skier) ] TJ -17.27110 -9.46367 Td [ (mak) 9.98605 (es) -203.00200 (his) -203.99400 (w) 10.01360 (ay) -203.01100 (through) -202.99900 (the) -203.98500 (sno) 25.01410 (w) ] TJ /R15 7.97010 Tf 114.00300 9.46367 Td (Ours) Tj /R17 7.97010 Tf 17.27110 0 Td [ (\072) -386.98700 (a) -289.01400 (skier) -289.00100 (is) -287.98400 (headed) -288.98900 (do) 25.02030 (wn) -289.02000 (a) ] TJ -17.27110 -9.46367 Td [ (steep) -250 (slope) ] TJ ET Q 2.99000 w 0 G 3179.66000 3601.46000 m 5360.09000 3601.46000 l S q 10 0 0 10 0 0 cm BT /R15 7.97010 Tf 1 0 0 1 338.40800 350.58300 Tm (Baseline) Tj /R17 7.97010 Tf 28.34100 0 Td [ (\072) -309.98700 (a) -250 (man) -250.01900 (riding) -250.01300 (skis) -250.01900 (do) 25.02030 (wn) -250.00700 (a) -250 (sno) 25.01410 (w) -249.99400 (co) 14.99130 (v) 14.98520 (ered) -249.98800 (slope) ] TJ ET Q 3179.66000 3458.79000 m 5360.09000 3458.79000 l S q 10 0 0 10 0 0 cm BT /R17 9.96260 Tf 1 0 0 1 308.86200 324.61000 Tm [ (Figure) -359.01400 (1\072) -528.01700 (F) 14.99260 (our) -358.99200 (images) -357.99700 (from) -358.98900 (the) -358.98900 (test) -358.98400 (set) -358.98900 (related) -358.99400 (to) -358.98900 (skiing\054) ] TJ 11.95510 TL T* [ (with) -284.98900 (capt) 0.98268 (ions) -284.98900 (from) -285.00100 (our) -283.98400 (model) -284.98700 (and) -283.98700 (a) -285.01600 (baseline\056) -413.01000 (Baseline) -284.98700 (de\055) ] TJ T* [ (scribes) -273.98100 (all) -273.00300 (four) -273.99100 (images) -274.01100 (with) -274.00800 (a) -273.99800 (gene) 0.98023 (ric) -273.99300 (caption\054) -280.00700 (whereas) -273.99300 (our) ] TJ T* [ (model) -250.01200 (produces) -249.98300 (di) 24.98850 (v) 14.98280 (erse) -250.00200 (and) -249.99300 (more) -249.98800 (image) -249.98300 <73706563690263> -250.01700 (captions\056) ] TJ 35.92770 TL T* [ (As) -205.01500 (we) -204.98800 (analyze) -206.00500 (in) -205.01200 (this) -205 (paper) 39.99330 (\054) -213.99900 (this) -206.02000 (is) -205.01000 (lik) 10.01790 (ely) -205.01500 (due) -205.02000 (to) -205.99500 (a) 1.01454 (rtif) 10.00810 (acts) -206 (and) ] TJ 11.95510 TL T* [ <6465026369656e63696573> -195.98400 (in) -195.99400 (the) -196 (statistics) -196.00200 (of) -196.00700 (the) -195.99700 (generated) -196.00700 (captions\054) -206.99500 (which) ] TJ T* [ (is) -239.98700 (more) -239.00900 (apparent) -239.99400 (when) -239.99200 (observing) -240.00900 (mult) 1.00964 (iple) -239.98400 (samples\056) -307.01700 (Specif\055) ] TJ T* [ (ically) 64.98670 (\054) -414.98700 (we) -382.99700 (observ) 14.99260 (e) -382.02000 (that) -382 (state\055of\055the\055art) -382.00500 (systems) -382.99500 (frequently) ] TJ T* [ (\223re) 25.00320 (v) 14.98280 (eal) -214.00600 (themselv) 15.01220 (es\224) -214.00600 (by) -213.01300 (generating) -213.98400 (a) -214.00800 (dif) 24.98600 (ferent) -214.00800 (w) 10.00320 (ord) -213.99900 (distrib) 19.99180 (u\055) ] TJ T* [ (tion) -278.98800 (and) -279.00700 (using) -278.98300 (smaller) -278.99300 (v) 20.00160 (ocab) 19.99670 (ulary) 65.01860 (\056) -398.00700 (Further) -278.99300 (scrutin) 15.00240 (y) -279.00700 (re) 25.00560 (v) 14.98280 (eals) ] TJ 11.95630 TL T* [ (that) -290.99600 (generalization) -290.98600 (from) -291 (the) -289.98100 (training) -290.98100 (set) -291 (is) -290.99600 (still) -291.01500 (challenging) ] TJ 11.95510 TL T* [ (and) -211.01900 (generation) -211.98700 (is) -211.01100 (biased) -211.00900 (to) -211.99400 (frequent) -211.01900 (fragments) -211.00400 (and) -211.99900 (captions\056) ] TJ 11.95510 -14.17700 Td [ (Also\054) -322.98000 (today\047) 54.98590 (s) -309.00200 (systems) -309.00700 (are) -307.99800 (e) 25.01050 (v) 24.98110 (aluated) -308.99700 (to) -308.01700 (produce) -308.99700 (a) -309.01200 (single) ] TJ -11.95510 -11.95590 Td [ (caption\056) -347.99800 (Y) 99.98580 (et\054) -265.01000 (multiple) -263.00800 (potentially) -261.98600 (distinct) -263.01500 (captions) -262.01000 (are) -262.98600 (typi\055) ] TJ T* [ (cally) -237.00900 (correct) -236.98700 (for) -236.98300 (a) -238.00400 (single) -236.99200 (image) -237.00500 (\226) -237.01400 (a) -236.98500 (property) -237.00500 (that) -237.98500 (is) -237.00500 <7265036563746564> ] TJ T* [ (in) -277.98100 (human) -278.01000 (ground\055truth\056) -392.99100 (This) -278.00800 (di) 24.98600 (v) 14.98280 (ersity) -278.01200 (is) -278.01700 (not) -278.01200 (equally) -278.01200 (repro\055) ] TJ T* [ (duced) -249.99000 (by) -249.98800 (state\055of\055the\055art) -249.98500 (caption) -250.01700 (generators) -249.98800 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R17 9.96260 Tf 1 0 0 1 488.40800 107.13300 Tm (40) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R17 9.96260 Tf 1 0 0 1 498.37000 107.13300 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R17 9.96260 Tf 1 0 0 1 503.35200 107.13300 Tm (23) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R17 9.96260 Tf 1 0 0 1 513.31400 107.13300 Tm (\135\056) Tj -192.49700 -14.17770 Td [ (Therefore\054) -212.01100 (our) -203.01800 (goal) -201.99300 (is) -203.01300 (to) -201.99600 (mak) 10.01300 (e) -202.99100 (image) -203.01000 (captions) -202.02000 (less) -203.01000 (distin\055) ] TJ -11.95510 -11.95510 Td [ (guishable) -194.99400 (from) -195.99700 (human) -195.00400 (ones) -195.01600 (\226) -196.00200 (similar) -194.99200 (in) -195.01600 (the) -195.99700 (spirit) -194.99700 (to) -195.01600 (a) -196.01100 (T) 45.01700 (uring) ] 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 (4135) 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 486
/Length 30626
/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 ? >Q 0}ip;?
kA=s@;R`x^h :S4 zәAHTc!\6=sR4N=>e9cH>JJAsLBp@lrj@
@SӎO 1jrzYjtlLB6KSm98=釿F =r}@)í 4A)Qך1bg AFOzc}if2qRzSqaNp?638 $fbsF3sR doZ $9&xv(-)Ry!BD*F;ӋH8ϵd~<.
U4qA@ǎ;Jpj-bG NGZn9'n@P
)zTl9C03NZ@.֝I
PGc@IP&=4Zw0dQG^hU=iM*apsB`$t4A9n9h Fi :\9 f{ JSѴRZ fݼӕ9敇j#ph+FxQG&6#$#0Gjb8-9TF-qނI :zsHaoJvܿ>å !b ]N1G4y R&qjiqӧ0nfg9$2Ub*A4i x u)'>13zP9G R@)~43@zR0~gkښ3RcJ`0F':T=1@,ةRzi5Td\@e&HUH'4u^ #01\.ø}