%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 (End\055To\055End Instance Segmentation With Recurrent Attention) /Producer (PyPDF2) /Author (Mengye Ren\054 Richard S\056 Zemel) /Subject (2017 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 >> /ExtGState << /s9 26 0 R /s11 29 0 R /a0 << /CA 1 /ca 1 >> /R21 32 0 R /s5 33 0 R /s7 36 0 R >> /Font << /F2 39 0 R /R28 40 0 R /F1 44 0 R /R34 45 0 R /R22 48 0 R /R32 51 0 R /R24 54 0 R /R30 58 0 R /R26 62 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 ] >> endobj 4 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 77 0 R /Resources << /XObject << /R49 78 0 R /R48 79 0 R /R47 80 0 R /R82 81 0 R /R81 82 0 R /R80 83 0 R /R50 84 0 R /R54 85 0 R /R55 86 0 R /R56 87 0 R /R78 88 0 R /R79 89 0 R /R58 90 0 R /R59 91 0 R /R72 92 0 R /R73 93 0 R /R70 94 0 R /R71 95 0 R /R76 96 0 R /R51 97 0 R /R74 98 0 R /R77 99 0 R /R52 100 0 R /R75 101 0 R /R53 102 0 R /R65 103 0 R /R64 104 0 R /R67 105 0 R /R66 106 0 R /R61 107 0 R /R60 108 0 R /R63 109 0 R /R62 110 0 R /R69 111 0 R /R68 112 0 R /R57 113 0 R >> /ExtGState << /R21 32 0 R >> /Font << /R93 114 0 R /R95 118 0 R /F2 122 0 R /R89 123 0 R /F1 127 0 R /R83 128 0 R /R91 131 0 R /R22 48 0 R /R87 136 0 R /R24 54 0 R /R85 139 0 R /R97 143 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 148 0 R 149 0 R 150 0 R 151 0 R 152 0 R ] >> endobj 5 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 153 0 R /Resources << /XObject << /R112 154 0 R /R113 155 0 R /R111 156 0 R /R105 157 0 R /R106 158 0 R >> /ExtGState << /R21 32 0 R >> /Font << /R109 159 0 R /R93 114 0 R /R107 162 0 R /R95 118 0 R /F2 166 0 R /R89 123 0 R /F1 167 0 R /R83 128 0 R /R91 131 0 R /R22 48 0 R /R24 54 0 R /R85 139 0 R /R97 143 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 168 0 R ] >> endobj 6 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 169 0 R /Resources << /ExtGState << /R21 32 0 R >> /Font << /F2 170 0 R /R93 114 0 R /R95 118 0 R /R127 171 0 R /R126 174 0 R /R89 123 0 R /F1 178 0 R /R85 139 0 R /R83 128 0 R /R91 131 0 R /R22 48 0 R /R24 54 0 R /R129 179 0 R /R97 143 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 184 0 R 185 0 R 186 0 R 187 0 R 188 0 R ] >> endobj 7 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 189 0 R /Resources << /ExtGState << /R21 32 0 R >> /Font << /R93 114 0 R /F2 190 0 R /R89 123 0 R /F1 191 0 R /R22 48 0 R /R32 51 0 R /R24 54 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 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 ] >> endobj 8 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 228 0 R /Resources << /ExtGState << /R21 32 0 R >> /Font << /F2 229 0 R /R22 48 0 R /R30 58 0 R /F1 230 0 R /R93 114 0 R /R95 118 0 R /R85 139 0 R /R24 54 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 231 0 R 232 0 R 233 0 R 234 0 R 235 0 R 236 0 R 237 0 R 238 0 R 239 0 R 240 0 R 241 0 R 242 0 R 243 0 R 244 0 R 245 0 R 246 0 R 247 0 R 248 0 R 249 0 R 250 0 R 251 0 R ] >> endobj 9 0 obj << /Parent 1 0 R /Rotate 0 /Contents 252 0 R /Resources << /XObject << /R228 253 0 R /R229 254 0 R /R239 255 0 R /R238 256 0 R /R237 257 0 R /R236 258 0 R /R235 259 0 R /R234 260 0 R /R233 261 0 R /R232 262 0 R /R231 263 0 R /R230 264 0 R /R226 265 0 R /R227 266 0 R >> /ExtGState << /R21 32 0 R >> /Font << /F2 267 0 R /F1 268 0 R /R22 48 0 R /R24 54 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /Group 269 0 R /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 288 0 R 289 0 R 290 0 R ] /Type /Page >> endobj 10 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 291 0 R /Resources << /XObject << /R264 292 0 R /R265 293 0 R /R266 294 0 R /R267 295 0 R /R260 296 0 R /R261 297 0 R /R262 298 0 R /R263 299 0 R /R259 300 0 R /R258 301 0 R /R255 302 0 R /R254 303 0 R /R257 304 0 R /R256 305 0 R /R253 306 0 R >> /ExtGState << /R21 32 0 R >> /Font << /R93 114 0 R /F2 307 0 R /F1 308 0 R /R34 45 0 R /R22 48 0 R /R24 54 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 309 0 R 310 0 R 311 0 R 312 0 R 313 0 R 314 0 R 315 0 R 316 0 R ] >> endobj 11 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 317 0 R /Resources << /ExtGState << /R21 32 0 R >> /Font << /F2 318 0 R /R34 45 0 R /F1 319 0 R /R24 54 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 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 379 0 R ] >> endobj 12 0 obj << /Type /Catalog /Pages 1 0 R >> endobj 13 0 obj << /Length 21119 >> stream q q q 0.10000 0 0 0.10000 0 0 cm /R21 gs 0 g q 10 0 0 10 0 0 cm BT /R22 14.34620 Tf 1 0 0 1 110.65400 675.06700 Tm [ (End\055to\055End) -249.99000 (Instance) -249.99600 (Segmentation) -250.00200 (with) -250.01300 (Recurr) 18 (ent) -250.00800 (Attention) ] TJ /R24 11.95520 Tf 103.22700 -37.85820 Td [ (Mengye) -250.00200 (Ren) ] TJ /R26 7.97010 Tf 61.42500 4.33906 Td [ (1) -0.30019 ] TJ /R24 11.95520 Tf 4.73203 -4.33906 Td [ (\054) -250.01200 (Richard) -250.00800 (S\056) -249.99300 (Zemel) ] TJ /R26 7.97010 Tf 89.98590 4.33906 Td [ (1) -0.30019 ] TJ /R28 7.97010 Tf 4.23438 0 Td [ (\073) -0.10109 ] TJ /R26 7.97010 Tf 2.35273 0 Td [ (2) -0.30019 ] TJ /R24 11.95520 Tf -239.57800 -18.28590 Td [ (Uni) 24.99460 (v) 14.98620 (ersity) -249.98900 (of) -250.01500 (T) 80.01470 (oronto) ] TJ /R26 7.97010 Tf 103.49600 4.33789 Td [ (1) -0.30019 ] TJ /R24 11.95520 Tf 4.73203 -4.33789 Td [ (\054) -250.01200 (Canadian) -250.01400 (Institute) -250.01600 (for) -249.99500 (Adv) 25.00790 (anced) -250.00400 (Research) ] TJ /R26 7.97010 Tf 208.19700 4.33789 Td [ (2) -0.30019 ] TJ /R30 8.96640 Tf -227.70200 -18.28590 Td [ (f) -0.90126 ] TJ /R32 8.96640 Tf 4.60820 0 Td (mren\054zemel) Tj /R30 8.96640 Tf 53.79800 0 Td [ (g) -0.90126 ] TJ /R32 8.96640 Tf 4.60781 0 Td (\100cs\056toronto\056edu) Tj /R22 11.95520 Tf -142.77500 -41.04610 Td (Abstract) Tj /R34 9.96260 Tf 0.98000 0 0 1 49.28520 544.05900 Tm [ (While) -239 (con) 40.99270 (volutional) -239.00200 (neur) 15.01600 (al) -238.99700 (networks) -239.01800 (have) -239.01600 (gained) -239.01100 (impr) 38.00890 (essive) ] TJ 1.02000 0 0 1 50.11210 532.10400 Tm [ (success) -262.00900 (r) 37.00180 (ecently) -261.98500 (in) -262.00300 (solving) -261.99300 (structur) 35.99580 (e) 1.01386 (d) -262.00300 (pr) 36.00900 (ediction) -262.00200 (pr) 44.00460 (oblems) ] TJ 0.98000 0 0 1 50.11210 520.14900 Tm [ (suc) 15.01290 (h) -255.01600 (as) -253.98700 (semantic) -255.00800 (se) 41.01580 (gmentation\054) -255.00700 (it) -254.98900 (r) 37.99760 (emains) -255 (a) -255.01700 (c) 15.02100 (halleng) 10.98260 (e) -255.00200 (to) -255.00200 (dif\055) ] TJ 1.00200 0 0 1 50.11210 508.19400 Tm [ (fer) 37.01350 (entiate) -249.01200 (individual) -248.98400 (object) -250.00500 (instance) 1.00517 (s) -250.00700 (in) -248.99000 (the) -249.00900 (scene) 14.99690 (\056) -308.98500 (Instance) ] TJ 1.01100 0 0 1 50.11210 496.23900 Tm [ (se) 40.01330 (gmentation) -246.98200 (is) -246.98800 (very) -246.98000 (important) -247.00400 (in) -246.01500 (a) -246.99000 (variety) -246.99100 (of) -246.98400 (applications\054) ] TJ 1.01900 0 0 1 50.11210 484.28300 Tm [ (suc) 15.01130 (h) -245.01200 (as) -244.98700 (autonomous) -245.01100 (driving) 9.99468 (\054) -244.98100 (ima) 10.01390 (g) 8.98102 (e) -243.98800 (captioning) 9.01348 (\054) -244.98300 (and) -245.01900 (vi) 0.99803 (sual) ] TJ 1.01700 0 0 1 50.11210 472.32800 Tm [ (question) -243.99500 (answering) 14.99500 (\056) -304.01700 (T) 90.00130 (ec) 14.98900 (hniques) -243.98400 (that) -243.99700 (combine) -243.99200 (lar) 36.98990 (g) 9.98301 (e) -243.99200 (gr) 14.98780 (aph\055) ] TJ 1.02000 0 0 1 50.11210 460.37300 Tm [ (ical) -365.01000 (models) -365.00500 (with) -365.00200 (low\055le) 14.98570 (vel) -366.00100 (vi) 1.01266 (sion) -365.99400 (have) -364.98900 (been) -364.98900 (pr) 44.00460 (oposed) -365.00500 (to) ] TJ 0.98000 0 0 1 50.11210 448.41800 Tm [ (addr) 38.00820 (ess) -245.98200 (this) -245.99400 (pr) 46.00320 (oblem\073) -248.98600 (howe) 14.99100 (ver) 114.00900 (\054) -247.99700 (we) -245.99600 (pr) 46.00320 (opose) -246.00200 (an) -246.01000 (end\055to\055end) -245.98700 (r) 37.99890 (e\055) ] TJ 11.95510 TL T* [ (curr) 38.00760 (ent) -254.98400 (neur) 15.01600 (al) -255 (network) -256.01100 (\050RNN\051) -254.98800 (ar) 38.00140 (c) 15.01980 (hitectur) 38.00390 (e) -255.00200 (with) -254.98200 (an) -256.01200 (attention) ] TJ T* [ (mec) 14.98790 (hanism) -216.00400 (to) -215.99200 (model) -215.98000 (a) -217.00700 (human\055l) 0.98273 (ik) 9.99358 (e) -215.99400 (counting) -216.98600 (pr) 46.00320 (ocess\054) -224.01500 (and) -216 (pr) 46.00320 (o\055) ] TJ 0.98100 0 0 1 50.11210 412.55200 Tm [ (duce) -255.99300 (detailed) -255.98200 (instance) -255.01600 (se) 39.98350 (gmentations\056) -318.98400 (The) -255.01900 (network) -255.98400 (is) -255.99400 (jointly) ] TJ 1.02000 0 0 1 50.11210 400.59700 Tm [ (tr) 15.01340 (ained) -289 (to) -287.98900 (sequentially) -288.98200 (pr) 44.00460 (oduce) -287.98500 (r) 36.00300 (e) 38.99290 (gions) -287.99300 (of) -288.98700 (inter) 35.98260 (est) -287.99700 (as) -289.00400 (well) ] TJ 1.00400 0 0 1 50.11210 388.64200 Tm [ (as) -249.00300 (a) -250.01100 (dominant) -248.98200 (object) -249.98900 (se) 39.99140 (gmentation) -250.00800 (within) -248.99200 (eac) 15.01590 (h) -250.01200 (r) 37.01490 (e) 39.99750 (gion\056) -309.98300 (The) ] TJ 0.99400 0 0 1 50.11210 376.68700 Tm [ (pr) 45.01750 (oposed) -250.00600 (model) -249.99200 (ac) 15.01160 (hie) 14.99800 (ves) -249.98700 (competitive) -250 (r) 36.98540 (esults) -250.00300 (on) -249.99800 (the) -250 (CVPPP) ] TJ 1 0 0 1 50.11210 364.73200 Tm (\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 53.98750 364.73200 Tm (27) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 63.95000 364.73200 Tm [ (\135\054) -249.98900 (KITTI) -250.00100 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 103.53100 364.73200 Tm (12) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 113.49400 364.73200 Tm [ (\135\054) -249.98900 (and) -249.98300 (Cityscapes) -250 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 189.31900 364.73200 Tm (8) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R34 9.96260 Tf 1 0 0 1 194.30000 364.73200 Tm [ (\135) -249.99100 (datasets\056) ] TJ /R22 11.95520 Tf -144.18800 -36.76020 Td [ (1\056) -249.99000 (Intr) 18.01460 (oduction) ] TJ /R24 9.96260 Tf 1.00200 0 0 1 62.06720 308.74500 Tm [ (Instance) -248.00600 (se) 15.00790 (gmentation) -247.98500 (is) -247.01900 (a) -247.99100 (fundamental) -247.98200 (computer) -248.00200 (vision) ] TJ -11.93120 -11.95590 Td [ (problem\054) -249.98800 (which) -250.00500 (aims) -250.98900 (to) -250.00700 (assign) -250.00100 (pix) 15.01160 (el\055le) 24.99240 (v) 15.01160 (el) -250.01200 (instance) -249.98300 (labeling) ] TJ 1.02000 0 0 1 50.11210 284.83400 Tm [ (to) -304.01900 (a) -304.01100 (gi) 24.99220 (v) 13.99890 (en) -304.00400 (image\056) -480.01900 (While) -303.98400 (the) -304.00400 (standard) -304.00200 (semantic) -303.98600 (se) 14.98930 (gmenta\055) ] TJ 11.95510 TL T* [ (tion) -258 (problem) -257.98200 (entails) -256.99500 (assigning) -258.00400 (class) -258.01600 (labels) -257.99500 (to) -258.00400 (each) -258.00200 (pix) 15.00370 (el) -257.99700 (in) ] TJ T* [ (an) -244.99800 (image\054) -243.99500 (it) -244.98100 (says) -243.99300 (nothing) -244.98700 (about) -243.98600 (the) -244.99800 (number) -244.98700 (of) -243.98300 (instances) -244.99700 (of) ] TJ T* [ (each) -311.01100 (class) -310.98700 (in) -311.01500 (the) -312 (image\056) -502.00700 (Instance) -310.99100 (se) 14.98930 (gmentation) -311.00200 (is) -311 (consid\055) ] TJ T* [ (erably) -298.99600 (more) -298.01000 (dif) 23.98440 <02> 0.99825 (cult) -299.00800 (than) -299 (semantic) -298.99000 (se) 14.98930 (gmentation) -299.00800 (because) ] TJ 11.95590 TL T* [ (it) -347.00200 (necessitates) -348.01200 (distinguishing) -346.99000 (nearby) -348 (and) -347.01300 (occluded) -348.01000 (object) ] TJ 11.95510 TL T* [ (instances\056) -653.01800 (Se) 15.00490 (gmenting) -361.98500 (at) -361.01800 (the) -362.01100 (instance) -360.99300 (le) 24.00120 (v) 14.99890 (el) -361.01700 (is) -362.00900 (useful) -360.99100 (for) ] TJ T* [ (man) 15.01820 (y) -268.99900 (tasks\054) -273.99300 (such) -267.99500 (as) -268.00800 (highlighting) -268.00100 (the) -268.98500 (outline) -268.01300 (of) -268.00800 (objects) -267.98200 (for) ] TJ 1.01600 0 0 1 50.11210 189.19300 Tm [ (impro) 14.99470 (v) 15.01820 (ed) -247.98100 (recognition) -247.00100 (and) -247.01000 (allo) 24.98810 (wing) -248.00300 (robots) -246.99400 (to) -246.99400 (delineate) -247.00600 (and) ] TJ 1.01900 0 0 1 50.11210 177.23800 Tm [ (grasp) -245.99700 (indi) 24.99630 (vidual) -245.98700 (objects\073) -245.01700 (it) -245.99900 (plays) -246.01200 (a) -244.98900 (k) 9.98145 (e) 14.00720 (y) -245.01100 (role) -246.00500 (in) -246.01700 (a) 1.00283 (utonomous) ] TJ 0.99800 0 0 1 50.11210 165.28300 Tm [ (dri) 25.00910 (ving) -251.01400 (as) -250.00500 (well\056) -311.00900 (Obtaining) -249.98800 (instance) -250.99800 (le) 24.99680 (v) 14.99320 (el) -251 (pix) 16 (el) -251 (labels) -250.99200 (is) -250.01500 (also) ] TJ 0.98000 0 0 1 50.11210 153.32700 Tm [ (an) -256 (important) -256.00700 (step) -256.99600 (to) 25.98480 (w) 9.99358 (ards) -256.00800 (general) -255.98600 (machine) -256.01000 (understanding) -256.98000 (of) ] TJ 1 0 0 1 50.11210 141.37200 Tm (images\056) Tj 1.00600 0 0 1 62.06720 129.11900 Tm [ (Instance) -247.99700 (se) 14.98720 (gmentation) -248.00800 (has) -247.01900 (been) -248.00200 (rapidly) -247.99700 (g) 5.01320 (aining) -247.99200 (in) -247.01100 (popu\055) ] TJ 1.01600 0 0 1 50.11210 117.16400 Tm [ (larity) 63.98140 (\054) -246.01900 (with) -246.98300 (a) -246.01900 (spurt) -245.99700 (of) -245.98500 (research) -246.98900 (papers) -246.01600 (in) -245.99100 (the) -246.99200 (past) -245.99300 (tw) 10.00730 (o) -246.01500 (years\054) ] TJ 0.99500 0 0 1 50.11210 105.20900 Tm [ (and) -249.99300 (a) -249.98200 (ne) 25.00880 (w) -249.99300 (benchmark) -250 (competition\054) -251.00900 (based) -250 (on) -250.00300 (the) -250.01700 (Cityscapes) ] TJ 1 0 0 1 50.11210 93.25310 Tm [ (dataset) -249.98500 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R24 9.96260 Tf 1 0 0 1 83.58630 93.25310 Tm (8) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R24 9.96260 Tf 1 0 0 1 88.56760 93.25310 Tm (\135\056) Tj 0.98000 0 0 1 62.06720 81 Tm [ (A) -241.98500 (sensible) -243.00800 (approach) -241.99100 (to) -241.99800 (instance) -242.98000 (se) 16.01000 (gmentation) -242.98700 (is) -242.01100 (to) -243 (formu\055) ] TJ 251.83200 487.26800 Td [ (late) -211.98700 (it) -212.01900 (as) -212.00400 (a) -211.99200 (structured) -211.98200 (output) -212.00400 (problem\056) -301.98500 (A) -212.01900 (k) 10.00730 (e) 15.01850 (y) -212.00400 (challenge) -211.98400 (here) -212.00900 (is) ] TJ 0.98400 0 0 1 308.86200 556.31300 Tm [ (the) -254.99600 (dimensionality) -254.98600 (of) -255.01400 (the) -254.99600 (structured) -255.99000 (out) 0.99119 (put\054) -256.01500 (which) -255.01400 (can) -254.99900 (be) -254.98900 (on) ] TJ 0.98000 0 0 1 308.86200 544.35700 Tm [ (the) -215.01500 (order) -214.98300 (of) -216.00500 (the) -215.01500 (number) -215.00300 (of) -215.00500 (pix) 14.98350 (els) -214.99000 (times) -215.99500 (the) -215.01500 (number) -215 (of) -215.00500 (objects\056) ] TJ 0.98800 0 0 1 308.86200 532.40200 Tm [ (Standard) -252.01700 (fully) -251.98500 (con) 40.01030 (v) 20.00640 (ol) 1 (utional) -253.00900 (netw) 11.00280 (orks) -252.98400 (\050FCN\051) -251.98200 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R24 9.96260 Tf 1 0 0 1 493.35700 532.40200 Tm (26) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R24 9.96260 Tf 0.98800 0 0 1 503.32000 532.40200 Tm [ (\135) -252.00200 (will) -251.98700 (ha) 20.00640 (v) 15.00600 (e) ] TJ 0.98000 0 0 1 308.86200 520.44700 Tm [ (trouble) -248.99800 (directly) -248.01000 (outputting) -249.01800 (all) -249.01300 (instance) -247.98300 (labels) -249.00300 (in) -248.99800 (a) -249.00300 (single) -248.01300 (shot\056) ] TJ 11.95470 TL T* [ (Recent) -249.98600 (w) 9.99483 (ork) -250.00800 (on) -249.01100 (instance) -249.98300 (se) 15.00850 (gmentation) -249.98800 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R24 9.96260 Tf 1 0 0 1 465.06400 508.49200 Tm (38) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R24 9.96260 Tf 0.98000 0 0 1 475.02600 508.49200 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R24 9.96260 Tf 1 0 0 1 479.90700 508.49200 Tm (45) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R24 9.96260 Tf 0.98000 0 0 1 489.86900 508.49200 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R24 9.96260 Tf 1 0 0 1 494.75000 508.49200 Tm (44) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R24 9.96260 Tf 0.98000 0 0 1 504.71300 508.49200 Tm [ (\135) -250.01800 (proposes) ] TJ 1.01300 0 0 1 308.86200 496.53700 Tm [ (comple) 15.00340 (x) -246.02000 (graphical) -246.98000 (models\054) -245.99100 (which) -245.98600 (results) -246.00500 (in) -247.01200 (a) -246.01000 (comple) 15.00340 (x) -246.02000 (and) ] TJ 0.98100 0 0 1 308.86200 484.58200 Tm [ (time\055consuming) -256.00400 (pipeline\056) -318.01700 (Furthermore\054) -255.98900 (these) -255.98400 (models) -256.00400 (cannot) ] TJ 1 0 0 1 308.86200 472.62600 Tm [ (be) -249.99700 (trained) -249.98500 (in) -249.98500 (an) -249.99700 (end\055to\055end) -249.98300 (f) 9.99343 (ashion\056) ] TJ 1.00600 0 0 1 320.81700 457.65100 Tm [ (One) -247.99700 (of) -247.00600 (the) -247.99200 (main) -248.01700 (challenges) -247.01100 (in) -247.98500 (instance) -248.00500 (se) 14.98600 (gmentation\054) -246.98200 (as) ] TJ 0.98000 0 0 1 308.86200 445.69600 Tm [ (in) -225.99500 (man) 15 (y) -226.01000 (o) -0.99773 (t) 1.01524 (her) -226.99000 (computer) -226.01300 (vision) -226.98300 (tasks) -225.99800 (such) -225.99800 (as) -227.00800 (object) -225.98300 (detection\054) ] TJ 0.99300 0 0 1 308.86200 433.74100 Tm [ (is) -251.98700 (occlusion\056) -313.99200 (F) 15.00940 (or) -251.98400 (a) -252.99900 (bottom\055up) -251.98700 (approach) -252.00700 (to) -251.98200 (handle) -253.00900 (occlusion\054) ] TJ 0.98200 0 0 1 308.86200 421.78500 Tm [ (it) -255.01400 (must) -254.98400 (sometimes) -255.00900 (mer) 19.01560 (ge) -254.99400 (tw) 9.98695 (o) -254.99400 (re) 15.01790 (gions) -254.98400 (that) -254.98900 (are) -255.01400 (not) -254.98900 (connected\054) ] TJ 1.01300 0 0 1 308.50300 409.83000 Tm [ (which) -246.99200 (becomes) -247.00200 (v) 15.01310 (ery) -247.00900 (challenging) -246.98800 (at) -248.00800 (a) -247.01700 (local) -246.98800 (scale\056) -306.98700 (M) 1.00636 (an) 14.01160 (y) -246.98800 (ap\055) ] TJ 0.99200 0 0 1 308.86200 397.87500 Tm [ (proaches) -252.01400 (to) -253.01900 (handle) -251.99900 (occlusion) -253.01200 (utilize) -251.99400 (a) -252.99200 (form) -251.98900 (of) -252.99700 (non\055maximal) ] TJ 1.02000 0 0 1 308.86200 385.92000 Tm [ (suppression) -253.98700 (\050NMS\051\054) -254.00700 (which) -254.01400 (is) -253.99000 (typically) -253.00500 (dif) 23.98200 <02> 1 (cult) -253.99500 (to) -254.00900 (tune\056) -331 (In) ] TJ 0.98000 0 0 1 308.86200 373.96500 Tm [ (cluttered) -231.98900 (scenes\054) -236.00300 (NMS) -232 (may) -231.01400 (suppress) -232.01900 (the) -232.01900 (detection) -230.98900 (results) -231.98400 (for) ] TJ 1.02000 0 0 1 308.86200 362.01000 Tm [ (a) -255.99800 (hea) 19.98420 (vily) -256 (occluded) -255.98400 (object) -256.01300 (because) -255.99800 (it) -255.01300 (has) -256.00800 (too) -255.99800 (much) -256.01700 (o) 14.99650 (v) 13.99710 (erlap) ] TJ -0.35195 -11.95630 Td [ (with) -245.99900 (fore) 14.99410 (ground) -245.99400 (objects\056) -307.01600 (One) -245.98500 (moti) 23.99640 (v) 24.99100 (ation) -245.99400 (of) -246.97900 (this) -245.98900 (w) 10.00890 (ork) -246.01300 (is) -245.99400 (to) ] TJ 1.01300 0 0 1 308.86200 338.09900 Tm [ (introduce) -245.01400 (an) -245.00900 (iterati) 23.99530 (v) 15.01310 (e) -245.00400 (procedure) -244.98500 (to) -245 (perform) -246.01000 (dynamic) -244.98900 (NMS\054) ] TJ 1 0 0 1 308.86200 326.14400 Tm [ (reasoning) -250.01500 (about) -250.02000 (occlusion) -250.00700 (in) -249.98800 (a) -250.00200 (top\055do) 25 (wn) -249.98800 (manner) 54.98100 (\056) ] TJ 1.02000 0 0 1 320.81700 311.16900 Tm [ (A) -404.99300 (related) -404.00800 (problem) -405.01500 (of) -405.00700 (interest) -404.99800 (entails) -403.98900 (counting) -404.99800 (the) -404.98800 (in\055) ] TJ -11.72070 -11.95630 Td [ (stances) -377.00600 (of) -376.98700 (an) -377.00100 (object) -376.98200 (class) -376.98900 (in) -377.01800 (an) -377.00400 (image\056) -700.01300 (On) -377.00400 (its) -377.00400 (o) 23.99160 (wn) -377.00400 (this) ] TJ 11.95470 TL T* [ (problem) -298.99600 (is) -298.00600 (also) -298.98400 (of) -298.99100 (practical) -298.01400 (v) 23.99160 (alue\056) -464.99400 (F) 14.99170 (or) -298.98900 (instance\054) -311.99600 (counting) ] TJ 0.98000 0 0 1 308.86200 275.30300 Tm [ (pro) 15.01350 (vides) -236.98800 (useful) -236.98500 (population) -237.01500 (estimates) -237.00500 (in) -236.99500 (medical) -236.99500 (imaging) -237.02000 (and) ] TJ 1.02000 0 0 1 308.86200 263.34800 Tm [ (aerial) -264.01300 (imaging\056) -362.00900 (General) -263.99400 (object) -264.00800 (counting) -263.99900 (is) -263.98400 (fundamental) -264.98400 (to) ] TJ 1.01000 0 0 1 308.86200 251.39300 Tm [ (image) -247.99800 (understanding\054) -247.98800 (and) -248.98300 (our) -247.98300 (basic) -248.01200 (arithmetic) -247.98800 (intelligence\056) ] TJ 1.02000 0 0 1 308.86200 239.43800 Tm [ (Studies) -298.01800 (in) -296.98300 (applications) -298.01600 (such) -298.01800 (as) -296.99000 (image) -297.99000 (question) -298.00400 (answering) ] TJ 1.00300 0 0 1 308.86200 227.48200 Tm (\133) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R24 9.96260 Tf 1 0 0 1 312.18900 227.48200 Tm (1) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R24 9.96260 Tf 1.00300 0 0 1 317.17100 227.48200 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R24 9.96260 Tf 1 0 0 1 322.16200 227.48200 Tm (34) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R24 9.96260 Tf 1.00300 0 0 1 332.12400 227.48200 Tm [ (\135) -249 (re) 24.01220 (v) 14.98680 (eal) -249.00800 (that) -249.01000 (counting\054) -249.99700 (especially) -248.99600 (on) -250.00200 (e) 25.01390 (v) 14.98680 (eryday) -249.00500 (objects\054) ] TJ 0.98800 0 0 1 308.86200 215.52700 Tm [ (is) -252.98700 (a) -252.99400 (v) 15.00600 (ery) -252.98900 (challenging) -253.00200 (task) -252.98200 (on) -252.98400 (its) -252.98400 (o) 25.00670 (wn) -252.98400 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R24 9.96260 Tf 1 0 0 1 457.97200 215.52700 Tm (7) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R24 9.96260 Tf 0.98800 0 0 1 462.95300 215.52700 Tm [ (\135\056) -314.01000 (Counting) -253 (has) -252.98400 (been) ] TJ 0.99800 0 0 1 308.86200 203.57200 Tm [ (formulated) -251.99600 (in) -252.00600 (a) -251 <7461736b2d73706563690263> -252.00100 (setting\054) -251.98600 (either) -251.99100 (by) -251.98200 (detection) -252.00600 (f) 1.01657 (o) -0.99693 (l) 0.98219 (\055) ] TJ 1.00700 0 0 1 308.86200 191.61700 Tm [ (lo) 25.00460 (wed) -249.01900 (by) -249.00700 (re) 15.01010 (gression\054) -249.99500 (or) -249.00900 (by) -249.00900 (learning) -248.98000 (discriminati) 25.01190 (v) 15.00520 (ely) -248.98500 (with) -249 (a) ] TJ 1 0 0 1 308.86200 179.66200 Tm [ (counting) -250.00200 (error) -250.00700 (metric) -249.98300 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R24 9.96260 Tf 1 0 0 1 399.33200 179.66200 Tm (22) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R24 9.96260 Tf 1 0 0 1 409.29500 179.66200 Tm (\135\056) Tj 0.98000 0 0 1 320.81700 164.68600 Tm [ (T) 81.99660 (o) -203.00500 (tackle) -203.00500 (these) -203.00700 (challenges\054) -213.01500 (we) -202.99700 (propose) -202.98700 (a) -202.99200 (ne) 25.98610 (w) -203.01700 (model) -202.99200 (based) ] TJ 1.02000 0 0 1 308.86200 152.73100 Tm [ (on) -358.98200 (a) -360.01800 (recurrent) -359.00400 (neural) -359.00200 (netw) 8.99023 (ork) -358.99000 (\050RNN\051) -359 (that) -359.01400 (utilizes) -360.01800 (visual) ] TJ 11.95470 TL T* [ (attention\054) -311.99100 (to) -298.98200 (perform) -298.98200 (instance) -298.98900 (se) 14.99170 (gmentation\056) -466.00700 (W) 79.01410 (e) -299.01300 (consider) ] TJ 1.00100 0 0 1 308.86200 128.82100 Tm [ (the) -248.98900 (problem) -249.01600 (of) -248.99100 (counti) 0.99394 (ng) -249.01300 (jointly) -249.01300 (with) -249.00800 (instance) -249.00400 (se) 15.01190 (gmentation\056) ] TJ 1.02000 0 0 1 308.86200 116.86600 Tm [ (Our) -287.98300 (system) -288.00200 (addresses) -288.99700 (the) -288.01400 (dimensionality) -287.98500 (issue) -288.00900 (by) -287.98100 (using) -288.99000 (a) ] TJ 11.95590 TL T* [ (temporal) -299.98800 (chain) -299.98600 (that) -300.00800 (outputs) -300.01700 (a) -300.01200 (single) -299.01800 (instance) -299.98800 (at) -300.01200 (a) -300.01200 (time\056) -469.00600 (It) ] TJ 0.98900 0 0 1 308.86200 92.95510 Tm [ (also) -252.01000 (performs) -251.01100 (d) -0.99609 (yna) 0.99361 (mic) -252.00200 (NMS\054) -251.99500 (using) -251.99500 (an) -252.02000 (object) -251.00400 (that) -252 (is) -252.01500 (already) ] TJ 0.98000 0 0 1 308.86200 81 Tm [ (se) 15.00850 (gmented) -247.98000 (to) -249 (aid) -248.98300 (in) -249 (the) -248.98300 (disco) 15.99370 (v) 15.00850 (ery) -248.99800 (of) -249.01300 (an) -247.99800 (occluded) -249.01800 (object) -248.98800 (later) ] TJ 1 0 0 1 295.12100 51.11210 Tm (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 (6656) 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