%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 (Paying Attention to Descriptions Generated by Image Captioning Models) /Producer (PyPDF2) /Author (Hamed R\056 Tavakoli\054 Rakshith Shetty\054 Ali Borji\054 Jorma Laaksonen) /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 /x6 18 0 R /x12 21 0 R /x10 24 0 R >> /ExtGState << /s9 27 0 R /s11 30 0 R /a0 << /CA 1 /ca 1 >> /s5 33 0 R /s7 36 0 R /R31 39 0 R >> /Font << /R40 40 0 R /F2 44 0 R /F1 45 0 R /R38 46 0 R /R36 51 0 R /R34 55 0 R /R32 59 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 62 0 R 63 0 R 64 0 R 65 0 R 66 0 R 67 0 R 68 0 R 69 0 R 70 0 R 71 0 R 72 0 R 73 0 R 74 0 R 75 0 R 76 0 R 77 0 R 78 0 R 79 0 R 80 0 R 81 0 R 82 0 R 83 0 R ] >> endobj 4 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 84 0 R /Resources << /ExtGState << /R31 39 0 R >> /Font << /F2 85 0 R /R34 55 0 R /F1 86 0 R /R32 59 0 R /R40 40 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 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 ] >> endobj 5 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 128 0 R /Resources << /XObject << /R109 129 0 R /R108 130 0 R /R118 131 0 R /R112 132 0 R /R113 133 0 R /R110 134 0 R /R111 135 0 R /R116 136 0 R /R117 137 0 R /R115 138 0 R >> /ExtGState << /R31 39 0 R >> /Font << /R119 140 0 R /R40 40 0 R /F2 145 0 R /F1 146 0 R /R121 147 0 R /R34 55 0 R /R32 59 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] /ColorSpace << /R114 139 0 R >> >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 ] >> endobj 6 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 164 0 R /Resources << /ExtGState << /R31 39 0 R /R152 165 0 R /R137 166 0 R >> /Font << /R144 167 0 R /R146 171 0 R /R141 174 0 R /R153 181 0 R /R150 185 0 R /R148 188 0 R /R40 40 0 R /F2 192 0 R /F1 193 0 R /R121 147 0 R /R138 194 0 R /R34 55 0 R /R32 59 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 << /XObject << /R192 206 0 R /R193 207 0 R /R190 208 0 R /R191 209 0 R /R189 210 0 R >> /ExtGState << /R31 39 0 R /R152 165 0 R /R137 166 0 R >> /Font << /R181 211 0 R /R183 215 0 R /R185 219 0 R /R194 223 0 R /R40 40 0 R /F2 227 0 R /F1 228 0 R /R121 147 0 R /R179 229 0 R /R187 233 0 R /R150 185 0 R /R175 237 0 R /R169 241 0 R /R177 245 0 R /R171 249 0 R /R173 253 0 R /R34 55 0 R /R153 181 0 R /R32 59 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 257 0 R 258 0 R 259 0 R 260 0 R 261 0 R 262 0 R 263 0 R 264 0 R 265 0 R 266 0 R ] >> endobj 8 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 267 0 R /Resources << /ExtGState << /R31 39 0 R >> /Font << /R153 181 0 R /R150 185 0 R /R40 40 0 R /F2 268 0 R /F1 269 0 R /R34 55 0 R /R32 59 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 270 0 R 271 0 R 272 0 R 273 0 R 274 0 R 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 ] >> endobj 9 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 288 0 R /Resources << /XObject << /R239 289 0 R /R236 292 0 R /R243 293 0 R /R240 294 0 R /R248 295 0 R /R247 296 0 R /R244 297 0 R /R251 298 0 R /R250 299 0 R /R242 300 0 R /R241 301 0 R >> /ExtGState << /R137 166 0 R /R31 39 0 R /R238 290 0 R >> /Font << /R256 302 0 R /R153 181 0 R /R150 185 0 R /R40 40 0 R /F2 306 0 R /F1 307 0 R /R258 308 0 R /R121 147 0 R /R254 311 0 R /R34 55 0 R /R245 314 0 R /R32 59 0 R /R252 318 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 ] >> endobj 10 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 333 0 R /Resources << /ExtGState << /R31 39 0 R >> /Font << /R40 40 0 R /F2 334 0 R /R279 335 0 R /F1 338 0 R /R34 55 0 R /R32 59 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 ] >> endobj 11 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 353 0 R /Resources << /ExtGState << /R31 39 0 R >> /Font << /F2 354 0 R /R34 55 0 R /F1 355 0 R /R40 40 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ ] >> endobj 12 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 356 0 R /Resources << /ExtGState << /R31 39 0 R >> /Font << /F2 357 0 R /R34 55 0 R /F1 358 0 R /R40 40 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ ] >> endobj 13 0 obj << /Type /Catalog /Pages 1 0 R >> endobj 14 0 obj << /Length 21066 >> stream q q q 0.10000 0 0 0.10000 0 0 cm /R31 gs 0 g q 10 0 0 10 0 0 cm BT /R32 14.34620 Tf 1 0 0 1 73.33280 675.06700 Tm [ (P) 10.01330 (aying) -250.00800 (Attention) -250.01100 (to) -250.01200 (Descriptions) -250.00500 (Generated) -249.99800 (by) -250.01300 (Image) -250.00500 (Captioning) -249.98800 (Models) ] TJ /R34 11.95520 Tf 45.47730 -37.85820 Td [ (Hamed) -249.98400 (R\056) -249.98700 (T) 80.01370 (a) 19.98840 (v) 24.98440 (ak) 10.01150 (oli) ] TJ /R36 7.97010 Tf 92.35270 4.33906 Td [ (y) -0.19911 ] TJ /R34 11.95520 Tf 19.20630 -4.33906 Td [ (Rakshith) -250.00600 (Shetty) ] TJ /R38 7.97010 Tf 76.05780 4.33906 Td [ (\077) -0.30019 ] TJ /R34 11.95520 Tf 19.67620 -4.33906 Td [ (Ali) -250.00400 (Borji) ] TJ /R36 7.97010 Tf 42.84690 4.33906 Td [ (z) -0.19911 ] TJ /R34 11.95520 Tf 19.20590 -4.33906 Td [ (Jorma) -250.01000 (Laaksonen) ] TJ /R36 7.97010 Tf 83.99610 4.33906 Td [ (y) -0.20217 ] TJ -307.23200 -14.05820 Td [ (y) -0.19911 ] TJ /R34 11.95520 Tf 4.26211 -4.33789 Td [ (Dept\056) -310.00900 (of) -250.01400 (Computer) -250.01400 (Science\054) -249.99300 (Aalto) -249.98900 (Uni) 24.99570 (v) 14.98510 (ersity) 64.98870 (\054) -250.01200 (Finland\056) ] TJ /R38 7.97010 Tf -20.31410 -9.60977 Td [ (\077) -0.29866 ] TJ /R34 11.95520 Tf 4.73203 -4.33828 Td [ (Max) -250.00500 (Planck) -250 (Institute) -250.01600 (for) -249.99500 (Informatics\054) -249.99500 (Saarbr) ] TJ 212.83000 0.06016 Td (\250) Tj -0.99805 -0.06016 Td [ (uck) 10 (en\054) -249.99700 (German) 14.98920 (y) 65.00910 (\056) ] TJ /R36 7.97010 Tf -246.71800 -9.71875 Td [ (z) -0.19911 ] TJ /R34 11.95520 Tf 4.26094 -4.33906 Td [ (Dept\056) -310.00900 (of) -250.01500 (Computer) -250.01400 (Science\054) -249.99300 (Uni) 24.99570 (v) 14.98510 (ersity) -249.98900 (of) -250.01400 (Central) -249.98700 (Florida\054) -249.98700 (Orlando\054) -249.99300 (USA\056) ] TJ /R32 11.95520 Tf 23.01990 -41.04610 Td (Abstract) Tj /R40 9.96260 Tf -83.92770 -23.91020 Td [ (T) 91.99870 (o) -273.98900 (bridg) 9.98730 (e) -273.99800 (the) -273.98600 (gap) -274.01800 (between) -274.01600 (humans) -274.00300 (and) -274.01800 (mac) 14.98030 (hines) -274.01300 (in) -273.98100 (im\055) ] TJ -11.95510 -11.95470 Td [ (a) 10.00320 (g) 10.00320 (e) -293.01400 (under) 10.01490 (standing) -291.98400 (and) -292.99500 (describing) 9.99711 (\054) -303.98600 (we) -291.98200 (need) -293.00700 (further) -293.01700 (insight) ] TJ 11.95510 TL T* [ (into) -302.00400 (how) -302.00500 (people) -302.01500 (describe) -302.01200 (a) -301.00400 (per) 36.98160 (ceived) -301.99400 (scene) 15.01470 (\056) -465.99800 (In) -301.98600 (this) -302.00400 (paper) 111.01800 (\054) ] TJ 11.95590 TL T* [ (we) -293.98100 (study) -294.01000 (the) -293.98200 (a) 10.00320 (gr) 36.98520 (eement) -293.98800 (between) -294.99200 (bottom\055up) -293.99000 (saliency\055based) ] TJ 11.95510 TL T* [ (visual) -411.00300 (attention) -410.98500 (and) -411.01500 (object) -411.01400 (r) 37.01960 (eferr) 14.98890 (als) -411.99100 (i) 0.98758 (n) -412.00500 (scene) -410.99300 (description) ] TJ T* [ (constructs\056) -444.98200 (W) 91.98590 (e) -295.01400 (in) 40.00560 (vestigate) -294.98500 (the) -295.00200 (pr) 44.98390 (operties) -294.98300 (of) -294.99700 (human\055written) ] TJ T* [ (descriptions) -235.00800 (and) -233.98500 (mac) 14.98160 (hine\055g) 10.01300 (ener) 15.01960 (ated) -235.00700 (ones\056) -304.99300 (W) 91.98710 (e) -234.00500 (then) -235.00700 (pr) 44.98510 (opose) ] TJ T* [ (a) -311.98200 (saliency\055boosted) -311.98300 (ima) 10.01300 (g) 10.00320 (e) -311.99200 (c) 1.01454 (a) -1.01454 (pt) 0.99248 (ioning) -311.99200 (model) -312.01200 (in) -312.01400 (or) 36.98650 (der) -312.01900 (to) -312.01400 (in\055) ] TJ 11.95590 TL T* [ (vestigate) -324 <62656e65027473> -324.00500 (fr) 44.98640 (om) -324.01700 (low\055le) 15.00240 (vel) -323.98600 (cues) -323.98300 (in) -324.99200 (langua) 9.99098 (g) 10.00320 (e) -323.99000 (models\056) ] TJ 11.95510 TL T* [ (W) 91.98650 (e) -244.00300 (learn) -244.99900 (that) -244.01300 (\0501\051) -244 (humans) -244.98900 (mention) -244.01000 (mor) 36.98770 (e) -244.00300 (salient) -244.00300 (objects) -244.99600 (ear) 20.00400 (\055) ] TJ T* [ (lier) -201.00900 (than) -201.00300 (less) -201.01000 (salient) -200.99200 (ones) -201.01300 (in) -201.01300 (their) -201.00400 (descriptions\054) -211.00900 (\0502\051) -200.98900 (the) -201.01800 (better) ] TJ T* [ (a) -190.98300 (captioning) -190 (model) -191.01100 (performs\054) -202.00800 (the) -191.02000 (better) -189.99000 (attention) -190.98300 (a) 10.00320 (gr) 36.98650 (eement) ] TJ T* [ (it) -312.01200 (has) -312.01000 (with) -311.01400 (human) -312.00300 (descriptions\054) -326.99000 (\0503\051) -311.98800 (the) -312.01900 (pr) 44.98510 (oposed) -312.00400 (saliency\055) ] TJ T* [ (boosted) -256.00700 (model\054) -258.01700 (compar) 36.99380 (ed) -255.99500 (to) -255.98400 (its) -256.01500 (baseline) -256.00900 (form\054) -258.01100 (does) -255.98400 (not) -256.01900 (im\055) ] TJ 11.95590 TL T* [ (pr) 44.98390 (o) 10.00320 (ve) -335.01200 <7369676e690263616e746c> 0.99616 (y) -335.00800 (on) -334.99300 (the) -334.01500 (MS) -334.99600 (COCO) -333.99100 (database) 9.99343 (\054) -356.01700 (indicating) ] TJ 11.95510 TL T* [ (e) 19.99240 (xplicit) -310.00200 (bottom\055up) -309.98700 (boosting) -309.98100 (does) -310.01400 (not) -310.01000 (help) -310.01400 (when) -310.00700 (the) -310.01900 (task) -310.01200 (is) ] TJ T* [ (well) -264.99100 (learnt) -264.98800 (and) -266.01900 (tuned) -264.99600 (on) -265.00500 (a) -265.01000 (data\054) -268.98900 (\0504\051) -265.01600 (a) -265.99000 (better) -264.99700 (g) 10.00320 (ener) 15.01960 (alization) ] TJ T* [ (is\054) -227.98500 (howe) 14.99690 (ver) 110.99900 (\054) -227.01800 (observed) -221.98700 (for) -221.98400 (the) -221.99500 (saliency\055boosted) -223.01700 (model) -221.98700 (on) -221.99200 (un\055) ] TJ T* [ (seen) -249.99500 (data\056) ] TJ /R32 11.95520 Tf 35.74100 TL T* [ (1\056) -249.99000 (Intr) 18.01460 (oduction) ] TJ /R34 9.96260 Tf 11.95510 -18.92890 Td [ (The) -315.00200 (recent) -315.01000 (adv) 24.98240 (ancements) -314.98800 (in) -314.99400 (machine) -314.98900 (learning\054) -330.98400 (together) ] TJ -11.95510 -11.95510 Td [ (with) -199.98400 (the) -201.01700 (increase) -200.00800 (in) -201.01300 (the) -199.99800 (a) 19.99180 (v) 24.98110 (ailable) -201.01200 (computational) -199.99100 (po) 24.98600 (wer) 39.98840 (\054) -210.98200 (ha) 19.99670 (v) 14.98280 (e) ] TJ 11.95510 TL T* [ (increased) -354.01700 (the) -354.01200 (interest) -353.99900 (in) -354.00700 (solving) -354.99800 (hi) 0.99248 (gh\055le) 24.98600 (v) 14.98280 (el) -354.99700 (problems) -353.98700 (such) ] TJ T* [ (as) -249.01500 (image\055captioning) -247.98700 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 136.40800 224.46200 Tm (14) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 146.37100 224.46200 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 151.34200 224.46200 Tm (61) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 161.30500 224.46200 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 166.27600 224.46200 Tm (9) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 171.25700 224.46200 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 176.22900 224.46200 Tm (13) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 186.19100 224.46200 Tm [ (\135\054) -249.01800 (scene) -248 (and) -249.01200 (video) -249 (under) 20.00400 (\055) ] TJ -136.07900 -11.95510 Td [ (standing) -210.98700 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 89.29490 212.50700 Tm (51) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 99.25740 212.50700 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 103.86000 212.50700 Tm (52) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 113.82300 212.50700 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 118.41500 212.50700 Tm (39) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 128.37800 212.50700 Tm [ (\135\054) -218.98400 (and) -211.99800 (visual) -210.99800 (question) -212.00600 (answering) -211.00400 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 261.01000 212.50700 Tm (37) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 270.97200 212.50700 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 275.57500 212.50700 Tm (2) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 280.55600 212.50700 Tm (\135\056) Tj -230.44400 -11.95470 Td [ (The) -221.99800 (main) -221.98200 (goal) -221.99000 (of) -222 (t) 0.98758 (hese) -222 (problems) -222.00800 (is) -221.98900 (an) -222 (inference) -222.01000 (which) -221.98700 (ends) ] TJ 11.95630 TL T* [ (in) -259.98400 (a) -261.02000 (human\055lik) 9.99833 (e) -260 (response\056) -340.98400 (The) -259.99100 (nature) -261.01000 (of) -259.99300 (such) -259.98400 (responses) -260.99300 (of\055) ] TJ 11.95470 TL T* [ (ten) -312.99900 (necessitates) -314.00300 (interaction) -313.01900 (between) -314.01900 (se) 25.01790 (v) 14.98280 (eral) -313.01200 (lo) 24.98850 (w\055le) 25.01790 (v) 14.98280 (el) -313.98400 (cog\055) ] TJ T* [ (niti) 25.00260 (v) 14.98280 (e) -351.98500 (tasks\054) -377.99300 (e\056g\056\054) -378.00700 (perception) -352.00100 (and) -352.01400 (sentence) -352.01500 (planning) -351.98500 (in) -352.98800 (de\055) ] TJ T* [ (scribing) -296.00100 (images\056) -448.99900 (Measuring) -296.00200 (the) -297.00200 (capability) -295.99700 (of) -295.98700 (a) -295.99500 (machine) -296.99200 (in) ] TJ T* [ (replicating) -349.01200 (such) -350.00800 (interactions) -349.00700 (is) -348.98700 (challenging\056) -608.01900 (Although) -350.01800 (the) ] TJ 11.95590 TL T* [ (tri) 24.98730 (vial) -449.98800 (assessment) -450.98700 (techniques) -450.01600 (f) 9.99466 (acilitate) -449.98400 (understanding) -451.01800 (the) ] TJ 11.95510 TL T* [ (a) 19.99240 (v) 14.98280 (erage) -331.99500 (performance) -331.00400 (of) -331.98300 (the) -330.99600 (algorithms\054) -352.01000 (we) -331.98600 (yet) -330.99600 (need) -331.98100 (more) ] TJ T* [ (detailed) -278.99400 (studies) -279.02000 (to) -279 (understand) -278.98500 <73706563690263> -278.99300 (properties) -278.99300 (of) -279.99000 (the) -279.00500 (e) 15.01220 (x\055) ] TJ T* [ (isting) -228.98300 (methods) -228.00500 (in) -229.00800 (comparison) -228.00700 (with) -228.99900 (a) -228.98600 (human) -228.01800 (baseline\056) -302.98900 (In) -229.01800 (the) ] TJ T* [ (domain) -288 (of) -287.00900 (image) -288.01400 (description) -287.01500 (by) -287.98100 (machines\054) -296.98500 (agreement) -287.98100 (with) ] TJ 258.75000 473.10000 Td [ (visual) -250.01000 (attention) -250.00200 (is) -249.98500 (one) -249.99300 (such) -249.98800 (case\056) ] TJ 11.95510 -12.64300 Td [ (There) -238.98200 (e) 15.01220 (xists) -238.00700 (v) 24.98110 (arious) -239.00400 (theories) -237.98000 (about) -239 (human) -238.01400 (sentence) -239.01400 (con\055) ] TJ -11.95510 -11.95510 Td [ (struction) -353.00500 (and) -354.01400 (formation\056) -619 (W) 49.99170 (undt\047) 54.99080 (s) -354.01400 (theory) -352.98500 (of) -353 (sentence) -354.01400 (pro\055) ] TJ T* [ (duction) -401.00200 (moti) 25.00810 (v) 24.98110 (ates) -402.00600 (the) -400.98500 (role) -400.98700 (of) -402.01100 (object\047) 54.98590 (s) -400.98700 (importance) -401.00700 (in) -402.00200 (sen\055) ] TJ 11.95590 TL T* [ (tence) -434.98600 (production\056) -866.99300 (He) -434.98900 (proposes) -435 (that\054) -481.97800 (in) -435.01600 (a) -436.01100 (free) -435 (w) 10.00320 (ord) -436 (po\055) ] TJ 11.95510 TL T* [ (sitioning) -401.01600 (scenario\054) -439.01500 (not) -401.99400 (bound) -401.00700 (by) -400.98200 (an) 15.01710 (y) -400.98700 (traditional) -402.01100 (rule\054) -439.02000 (the) ] TJ T* [ (w) 10 (ords) -265.00200 (follo) 24.99830 (w) -263.99300 (each) -264.98300 (other) -265.00500 (according) -263.98500 (to) -265.00500 (the) -263.98500 (de) 15.01710 (gree) -264.98000 (of) -265.01500 (empha\055) ] TJ T* [ (sis) -340.00200 (on) -340.01200 (the) -340.01400 (concepts) -338.98800 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 393.23600 469.72600 Tm (56) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 403.19800 469.72600 Tm [ (\135\056) -579.98200 (This) -339.99700 (theory) -340.00700 (implicitly) -340.00200 (moti) 25.00560 (v) 24.98110 (ates) ] TJ -94.33590 -11.95510 Td [ (the) -316.01800 (role) -315.98200 (of) -315.98400 (what) -316.01600 (is) -316.01300 (later) -315.98400 (on) -316.01600 (recognized) -316.02100 (as) -315.98600 (salienc) 14.99260 (y) 65.01370 (\056) -507.98900 (Grif) 25.02030 <026e> ] TJ 11.95470 TL T* [ (and) -352.01500 (Bock) -352 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 354.61100 445.81600 Tm (21) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 364.57300 445.81600 Tm [ (\135) -351.98300 (found) -352.00300 (some) -352.98000 (empirical) -351.99500 (supporting) -352.01000 (e) 25.01050 (vidence) ] TJ -55.71090 -11.95630 Td [ (by) -330.99400 (sho) 24.99340 (wing) -330.98400 (that) -330.98900 (while) -330.98600 (describing) -331.01300 (scenes\054) -352.01500 (speak) 10.00810 (ers) -331.00400 (look) -330.97900 (at) ] TJ 11.95470 TL T* [ (an) -210.98400 (object) -209.98900 (before) -210.98900 (naming) -210.99400 (it) -209.98900 (within) -210.98900 (their) -211.01400 (description\056) -297.00900 (Besides\054) ] TJ T* [ (there) -275.99300 (e) 15.01220 (xists) -276 (numerous) -276 (studies) -275.99800 (which) -276.99800 (ha) 19.99670 (v) 14.98280 (e) -275.99800 (utilized) -276.00300 (attention) ] TJ T* [ (to) -201.01300 (analyze) -200.98600 (human) -201.00300 (sentence) -200 (planning) -200.99100 (and) -201.02000 (construction) -200.98600 (in) -201.01500 (dif\055) ] TJ T* [ (ferent) -297.99200 (scenarios) -298.99700 (including) -297.98500 (scene) -299.00900 (description\056) -454.98300 (Most) -298 (of) -299.00900 (their) ] TJ T* [ <026e64696e6773> -285.98400 (pro) 14.98520 (vide) -285 (supporting) -285.98200 (e) 25.01050 (vidence) -285.98200 (that) -284.99700 (the) -285.98200 (sentence) -285.98700 (for) 19.99670 (\055) ] TJ 11.95590 TL T* [ (mation) -301.01600 (and) -301.98400 (attention) -301.99400 (corr) 1.01454 (elate) -301.98900 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 435.97500 362.12900 Tm (6) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 440.95600 362.12900 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 446.45500 362.12900 Tm (40) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 456.41800 362.12900 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 461.90700 362.12900 Tm (22) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 471.87000 362.12900 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 477.36900 362.12900 Tm (26) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 487.33100 362.12900 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 492.82100 362.12900 Tm (64) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 502.78300 362.12900 Tm [ (\135\056) -464.98100 (Encour) 20.00650 (\055) ] TJ -193.92100 -11.95510 Td [ (aged) -224 (enough\054) -228.99900 (we) -224.00400 (lay) -223.01400 (the) -223.99400 (foundation) -224.00200 (of) -224.00200 (this) -222.99700 (study) -224.01200 (on) -223.99200 (the) -223.99200 (role) ] TJ 11.95470 TL T* [ (of) -212.98200 (salienc) 14.99010 (y) -213.99900 (in) -213.01100 (the) -213.99600 (construct) -212.99900 (of) -212.98400 (image) -213.98900 (descriptions\054) -219.99800 (where) -214.00800 (the) ] TJ T* [ (order) -249.99700 (of) -249.99500 (named) -249.98300 (objects) -250.01500 (is) -249.98500 (momentous) -249.98300 (in) -249.98800 (a) -250.00200 (sentence\056) ] TJ /R32 9.96260 Tf 20.34610 TL T* [ (Contrib) 19.99420 (ution\072) ] TJ /R34 9.96260 Tf 68.99100 0 Td [ (In) -406.98800 (this) -405.98600 (paper) 39.99330 (\054) -446 (we) -406.99100 (address) -406.98100 (tw) 10.00810 (o) -406.98600 (intriguing) ] TJ -68.99100 -11.95510 Td [ (questions\072) -366.00800 (\0501\051) ] TJ /R40 9.96260 Tf 58.43590 0 Td [ (How) -278.01000 (well) -278.01000 (do) -277.98300 (ima) 10.01300 (g) 10.00320 (e) -277.01700 (descriptions\054) -284.99700 (by) -277.99300 (humans) ] TJ -58.43590 -11.95470 Td [ (or) -197.01400 (models\054) -207 (on) -195.99700 (a) -196.98200 (scene) -196.00900 (a) 10.00320 (gr) 36.98650 (ee) -196.99700 (with) -196.01100 (saliency) ] TJ /R34 9.96260 Tf 162.48400 0 Td [ (\077\054) -206.98500 (\0502\051) ] TJ /R40 9.96260 Tf 22.55310 0 Td [ (Can) -197.00100 (saliency) ] TJ -185.03700 -11.95630 Td [ <62656e650274> -249.98300 (ima) 10.01300 (g) 10.00320 (e) -250.00200 (captioning) -249.01000 (by) -249.99700 (mac) 14.98280 (hine) ] TJ /R34 9.96260 Tf 146.29300 0 Td [ (\077) -309.99200 (Answering) -250.01200 (the) -249.00700 (ques\055) ] TJ -146.29300 -11.95510 Td [ (tions\054) -379.00300 (we) -353.00200 (learn) -353 (not) -352.98300 (only) -353.99700 (about) -352.98000 (the) -352.99000 (role) -352.99500 (of) -353 (attention) -353.98500 (in) -352.99000 (de\055) ] TJ 11.95470 TL T* [ (scribing) -198.01800 (images\054) -209 (b) 20.00160 (ut) -197.99400 (also) -199.01100 (about) -197.98700 (the) -197.99600 (quality) -198.01100 (of) -198.98600 (human\055written) ] TJ T* [ (descriptions) -200.00300 (and) -200 (machine\055generated) -199.99100 (ones\056) -292.99500 (W) 79.98660 (e) -200.01100 <02727374> -199.98100 (study) -200.01600 (the) ] TJ T* [ (te) 14.98030 (xtual) -359.01900 (statistics) -358.99200 (of) -358.01600 (the) -358.99200 (sentences) -358.98900 (by) -358.98900 (human) -359.01400 (and) -358.99400 (machine\056) ] TJ T* [ (Then) -360.98900 (we) -361.00100 (in) 40.00560 (v) 14.98280 (estig) 5.00162 (ate) -361.00100 (the) -360.99100 (attention) -361.00300 (correlation) -362.00300 (in) -360.98900 (the) -360.98900 (struc\055) ] TJ 11.95590 TL T* [ (ture) -261.01000 (of) -259.99300 (human\055written) -261.01300 (and) -259.99100 (machine\055generated) -261 (descriptions\056) ] TJ 11.95510 TL T* [ (T) 79.99160 (o) -352.99500 (further) -352.99700 (e) 25.01050 (v) 24.98110 (aluate) -352.99300 (the) -352.99300 (contrib) 20.01630 (ution) -353.99000 (of) -353 (lo) 24.98600 (w\055le) 25.01540 (v) 14.98280 (el) -352.99500 (cues\054) -378.99100 (we) ] TJ T* [ (propose) -298.00900 (a) -297.99400 (salienc) 14.99010 (y\055boosted) -298 (captioning) -297.99000 (model) -298.98400 (and) -297.98500 (compare) ] TJ T* [ (it) -249.98300 (ag) 5.01877 (ainst) -250.01500 (a) -250.00200 (set) -249.98800 (of) -249.99500 (baseline) -250.01700 (captioning) -249.99700 (models\056) ] TJ /R32 11.95520 Tf 25.97380 TL T* [ (2\056) -249.98900 (Related) -250.00200 (W) 74.99720 (ork) ] TJ /R32 9.96260 Tf 19.61720 TL T* [ (Image) -209.98900 (description) -209.99200 (generation\056) ] TJ /R34 9.96260 Tf 136.18100 0 Td [ (There) -210.00900 (e) 15.01220 (xists) -210.01400 (a) -210.00900 (wide) -209.99400 (range) ] TJ -136.18100 -11.95470 Td [ (of) -382.99200 (captioning) -382.01700 (methods) -382.99700 (and) -382.99000 (models\056) -708.00400 (The) 14.98280 (y) -382.01000 (can) -383 (be) -382.99500 (cate) 15.01220 (go\055) ] TJ 11.95510 TL T* [ (rized) -416.98900 (into) -417.98400 (retrie) 25.01540 (v) 24.98110 (al\055based) -416.98400 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 418.76900 92.95550 Tm (19) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 428.73200 92.95550 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 435.37700 92.95550 Tm (44) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 445.33900 92.95550 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 451.99400 92.95550 Tm (25) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 461.95700 92.95550 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 468.60200 92.95550 Tm (33) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 478.56400 92.95550 Tm [ (\135\054) -458.98200 (sentence) -418.00400 (gen\055) ] TJ -169.70200 -11.95510 Td [ (eration) -398 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 343.81100 81.00040 Tm (34) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 353.77300 81.00040 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 360.21900 81.00040 Tm (16) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 370.18200 81.00040 Tm [ (\135\054) -434.98600 (and) -398.00700 (t) 0.98758 (he) -398.01200 (models) -397.98300 (which) -397.01700 (combine) -397.98800 (the) -398.00200 (tw) 10.00810 (o) ] 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 (2487) 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
/Resources <<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
/XObject <<
/x15 19 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
19 0 obj
<<
/Filter /FlateDecode
/Resources 20 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
20 0 obj
<<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
>>
endobj
21 0 obj
<<
/Filter /FlateDecode
/Resources <<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
/XObject <<
/x24 22 0 R
>>
>>
/Length 28
/Group <<
/Type /Group
/S /Transparency
/CS /DeviceRGB
/I true
>>
/BBox [ 133 751 479 772 ]
/Type /XObject
/Subtype /Form
>>
stream
x+O4PH/VЯ02Qp
0
endstream
endobj
22 0 obj
<<
/Filter /FlateDecode
/Resources 23 0 R
/Length 53008
/Type /XObject
/BBox [ 133 751 479 772 ]
/Subtype /Form
>>
stream
xlK8,8?DKs9mavd
{f-8*2Y@H >חk}y}uf`v)_s}1z#*Gw_gX jow\o'1c|Z^G<
A̺X}ay?IT|y~L.[ {Ȟb\3-3]_'X\竵0{+_۾oY-wj+ B;)Aa=/
1~a(>}m_K' >*Q1:?OPnGsQ_`?mzN
?WBB!lU n'S 1Ipzχ1xF_09,}?e){:̹~Gu7g {0mD`pdS)>;ys`Pµx'O
+혝 oI~!p7̵&m=dC+>5ALV"I1"S*CKha#03L4|@p灟vMRQ==*/m"r>@uqo9ͻ;Vz 'ȳ;IBzWR~ihwa8PlT>]fZ8y52)r빃W{B7G/q)}M]elS(k=5^A] сڸc'w!Ll'M=Ȥ뎓z;#կN>\q6r>|8}}Lr$Mo5:>syls+Ǵ`C?/9{Y||jD u#i^\rJ8a|oY+Rxl<{[؏s迨nYaյ[?/^u;+ ,bhL}wrj:ɴdM8^q|m'aQ6;0k_ar|E8HɃpN('&r`V;Rs볩Nk\&W=DDG#5]~M{VyCSRH9ܜXˏt7߾i9ADگ
USڝ|{(hbIRZXg @6Hɹ0Dq;si\Tv<1?mUŏV7bF
p7Y9sƨ*.uz_LբřsB݃ɹ^'_Yae91f+n%A̶d
Ċc#q
CF=Mt'\&ξ)z;L-4:n}41j:mӶ_)\݇ |T@uAƓxAN6Ql;}Ɲ{~_y"g%]nCi']I~,xa?feoɌr0u?V=phldx.Y;~nZL603> \`ęܓ