%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 (Active Object Localization With Deep Reinforcement Learning) /Producer (PyPDF2) /Author (Juan C\056 Caicedo\054 Svetlana Lazebnik) /Subject (2015 IEEE International Conference on Computer Vision) >> endobj 3 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 13 0 R /Resources << /XObject << /x8 14 0 R /R27 17 0 R /R25 18 0 R /x6 19 0 R /x12 22 0 R /x10 25 0 R /R26 28 0 R >> /ExtGState << /s9 29 0 R /s11 32 0 R /R16 35 0 R /a0 << /CA 1 /ca 1 >> /s5 36 0 R /s7 39 0 R >> /Font << /R29 42 0 R /F2 49 0 R /F1 50 0 R /R17 51 0 R /R21 54 0 R /R19 57 0 R /R23 61 0 R /R32 65 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 << /ExtGState << /R16 35 0 R >> /Font << /F1 78 0 R /R17 51 0 R /F2 79 0 R /R64 80 0 R /R66 83 0 R /R60 87 0 R /R62 92 0 R /R19 57 0 R /R23 61 0 R /R68 95 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 ] >> endobj 5 0 obj << /Parent 1 0 R /Rotate 0 /Contents 117 0 R /Resources << /XObject << /R102 118 0 R /R98 119 0 R /R99 121 0 R /R89 122 0 R /R88 123 0 R /R107 125 0 R /R90 126 0 R /R91 127 0 R /R92 128 0 R /R93 129 0 R /R87 130 0 R /R95 132 0 R /R85 133 0 R /R84 124 0 R /R109 134 0 R /R108 135 0 R /R86 131 0 R /R106 136 0 R /R101 138 0 R /R100 139 0 R /R103 140 0 R /R111 141 0 R /R105 137 0 R /R104 143 0 R /R114 144 0 R /R115 153 0 R /R94 146 0 R /R96 147 0 R /R97 120 0 R /R110 142 0 R >> /ExtGState << /R82 145 0 R /R16 35 0 R >> /Font << /R118 154 0 R /R112 148 0 R /F1 159 0 R /R120 160 0 R /R116 163 0 R /F2 166 0 R /R64 80 0 R /R66 83 0 R /R60 87 0 R /R17 51 0 R /R19 57 0 R /R68 95 0 R /R62 92 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /Group 152 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 167 0 R 168 0 R 169 0 R 170 0 R 171 0 R 172 0 R ] /Type /Page >> endobj 6 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 173 0 R /Resources << /ExtGState << /R16 35 0 R >> /Font << /R118 154 0 R /R140 174 0 R /R142 177 0 R /F1 181 0 R /R62 92 0 R /F2 182 0 R /R66 83 0 R /R134 183 0 R /R60 87 0 R /R136 187 0 R /R17 51 0 R /R138 191 0 R /R19 57 0 R /R68 95 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 195 0 R 196 0 R 197 0 R 198 0 R 199 0 R 200 0 R ] >> endobj 7 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 201 0 R /Resources << /ExtGState << /R16 35 0 R >> /Font << /F2 202 0 R /R19 57 0 R /F1 203 0 R /R60 87 0 R /R17 51 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 ] >> endobj 8 0 obj << /Parent 1 0 R /Rotate 0 /Contents 215 0 R /Resources << /XObject << /R181 216 0 R /R182 217 0 R /R185 218 0 R /R186 223 0 R >> /ExtGState << /R82 145 0 R /R16 35 0 R >> /Font << /F2 224 0 R /R183 219 0 R /F1 225 0 R /R179 226 0 R /R17 51 0 R /R19 57 0 R /R23 61 0 R /R177 229 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /Group 152 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 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 ] /Type /Page >> endobj 9 0 obj << /Parent 1 0 R /Rotate 0 /Contents 244 0 R /Resources << /XObject << /R264 245 0 R /R265 247 0 R /R266 249 0 R /R267 250 0 R /R260 248 0 R /R261 251 0 R /R262 252 0 R /R263 246 0 R /R220 253 0 R /R221 254 0 R /R222 255 0 R /R223 257 0 R /R268 259 0 R /R225 276 0 R /R201 277 0 R /R248 262 0 R /R249 263 0 R /R246 278 0 R /R247 295 0 R /R244 283 0 R /R245 284 0 R /R242 285 0 R /R243 286 0 R /R197 296 0 R /R229 297 0 R /R202 298 0 R /R204 258 0 R /R209 299 0 R /R199 300 0 R /R205 301 0 R /R230 302 0 R /R218 303 0 R /R208 309 0 R /R211 256 0 R /R210 310 0 R /R213 305 0 R /R212 306 0 R /R215 307 0 R /R214 308 0 R /R239 279 0 R /R219 319 0 R /R236 280 0 R /R235 281 0 R /R234 282 0 R /R233 320 0 R /R232 321 0 R /R231 324 0 R /R269 329 0 R /R259 260 0 R /R258 261 0 R /R257 264 0 R /R256 265 0 R /R250 266 0 R /R253 267 0 R /R200 330 0 R /R198 331 0 R /R224 323 0 R /R228 322 0 R /R203 304 0 R >> /ExtGState << /R82 145 0 R /R16 35 0 R >> /Font << /R206 311 0 R /R216 315 0 R /R237 287 0 R /R19 57 0 R /R226 325 0 R /F2 332 0 R /F1 333 0 R /R17 51 0 R /R254 268 0 R /R251 272 0 R /R240 291 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /Group 152 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 334 0 R 335 0 R 336 0 R 337 0 R 338 0 R ] /Type /Page >> endobj 10 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 339 0 R /Resources << /ExtGState << /R16 35 0 R >> /Font << /R283 340 0 R /R281 344 0 R /F2 347 0 R /F1 348 0 R /R17 51 0 R /R19 57 0 R /R23 61 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 349 0 R 350 0 R 351 0 R 352 0 R 353 0 R 354 0 R ] >> endobj 11 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 355 0 R /Resources << /ExtGState << /R16 35 0 R >> /Font << /F2 356 0 R /R19 57 0 R /R23 61 0 R /F1 357 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 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 380 0 R 381 0 R 382 0 R 383 0 R 384 0 R 385 0 R 386 0 R 387 0 R 388 0 R 389 0 R 390 0 R 391 0 R 392 0 R 393 0 R 394 0 R 395 0 R 396 0 R 397 0 R 398 0 R 399 0 R 400 0 R 401 0 R 402 0 R 403 0 R ] >> endobj 12 0 obj << /Type /Catalog /Pages 1 0 R >> endobj 13 0 obj << /Length 20147 >> stream q q q 0.10000 0 0 0.10000 0 0 cm /R16 gs 0 g q 10 0 0 10 0 0 cm BT /R17 14.34620 Tf 1 0 0 1 106.52300 675.06700 Tm [ (Acti) 10.00900 (v) 9.99625 (e) -249.99900 (Object) -249.99900 (Localization) -250.00500 (with) -250.01300 (Deep) -250.00800 (Reinf) 25.00590 (or) 17.98780 (cement) -250.01200 (Lear) 14.98930 (ning) ] TJ /R19 11.95520 Tf 33.88010 -37.85820 Td [ (Juan) -250.00500 (C\056) -249.98700 (Caicedo) ] TJ -56.32110 -13.94690 Td (Fundaci) Tj 39.51800 0.05898 Td (\264) Tj -0.99883 -0.05898 Td [ (on) -250.01200 (Uni) 24.99460 (v) 14.98620 (ersitaria) -249.99800 (K) 34.99600 (onrad) -250.01200 (Lorenz) ] TJ 12.97770 -13.94800 Td (Bogot) Tj 29.89300 0.05977 Td (\264) Tj -0.66289 -0.05977 Td [ (a\054) -249.98600 (Colombia) ] TJ /R21 8.96640 Tf -76.84800 -13.94800 Td (juanc\056caicedor\100konradlorenz\056edu\056co) Tj /R19 11.95520 Tf 275.08700 41.84300 Td [ (Sv) 15.00350 (etlana) -250.00600 (Lazebnik) ] TJ -59.62500 -13.94690 Td [ (Uni) 24.99570 (v) 14.98510 (ersity) -249.98900 (of) -250.01400 (Illinois) -250.00800 (at) -249.98900 (Urbana) -250.01000 (Champaign) ] TJ 62.69220 -13.94800 Td [ (Urbana\054) -249.99700 (IL\054) -249.99700 (USA) ] TJ /R21 8.96640 Tf -15.32110 -13.94800 Td (slazebni\100illinois\056edu) Tj /R17 11.95520 Tf -204.79900 -41.04610 Td (Abstract) Tj /R23 9.96260 Tf -83.92770 -23.92380 Td [ (W) 91.98650 (e) -309.99200 (pr) 36.98520 (esent) -310.01700 (an) -310.01700 (active) -309.00100 (detection) -310.00300 (model) -310.01200 (for) -310.01000 (localizing) -310.01400 (ob\055) ] TJ -11.95510 -11.95510 Td [ (jects) -330.01200 (in) -330.01100 (scenes\056) -549 (The) -330.00900 (model) -330.00700 (is) -328.99000 <636c6173732d73706563690263> -329.99600 (and) -330.00900 (allows) -330.01800 (an) ] TJ 11.95510 TL T* [ (a) 10.00320 (g) 10.00320 (ent) -428 (to) -427.01500 (focus) -427.98900 (attention) -428.00200 (on) -427.99700 (candidate) -426.99500 (r) 37.01830 (e) 39.98840 (gions) -428.01700 (for) -427.99000 (identi\055) ] TJ T* [ (fying) -415.00900 (the) -414.98200 (corr) 36.98650 (ect) -414.99100 (location) -414.01200 (of) -415.01700 (a) -414.98500 (tar) 36.99260 (g) 10.00320 (et) -414.98700 (object\056) -804.98600 (This) -414.99700 (a) 10.00320 (g) 10.00320 (ent) ] TJ 11.95590 TL T* [ (learns) -292.00500 (to) -292.99800 (deform) -292.00800 (a) -293.00500 (bounding) -291.98800 (box) -293.00400 (using) -292 (simple) -292.98500 (tr) 14.99010 (ansforma\055) ] TJ 11.95510 TL T* [ (tion) -429.00200 (actions\054) -474.00500 (with) -428.99300 (the) -430 (goal) -429.00400 (of) -429.01400 (determining) -428.98500 (the) -429.01900 (most) -429.98500 (spe\055) ] TJ T* [ <63690263> -319.98200 (location) -319.98800 (of) -320.99300 (tar) 36.99260 (g) 10.00320 (et) -319.98300 (objects) -320.00300 (following) -320.00300 (top\055down) -321.00300 (r) 37.01830 (eason\055) ] TJ T* [ (ing) 14.99500 (\056) -442.98200 (The) -294.01400 (pr) 44.98390 (oposed) -294.00700 (localization) -294.98600 (a) 10.00320 (g) 10.00320 (ent) -293.98300 (is) -294.01700 (tr) 14.99010 (ained) -294.01200 (using) -294.98000 (deep) ] TJ T* [ (r) 37.01960 (einfor) 37 (cement) -331.98100 (learning) 9.99466 (\054) -352.01700 (and) -330.98900 (e) 15.01220 (valuated) -331.99900 (on) -330.99400 (the) -332.01600 (P) 79.99160 (ascal) -332.00300 (V) 30 (OC) ] TJ T* [ (2007) -278.01200 (dataset\056) -395 (W) 91.98590 (e) -279.01700 (show) -278.00300 (that) -278.00800 (a) 10.00320 (g) 10.00320 (ents) -278.99900 (guided) -278.01000 (by) -277.99300 (the) -279.00500 (pr) 44.98510 (oposed) ] TJ 11.95590 TL T* [ (model) -255 (ar) 36.98520 (e) -254.98100 (able) -255.98400 (to) -255.00400 (localize) -254.98700 (a) -255.01100 (single) -256.01000 (instance) -254.98900 (of) -255.00400 (an) -255.00600 (object) -255.98200 (af\055) ] TJ 11.95510 TL T* [ (ter) -285.00200 (analyzing) -285.00900 (only) -285 (between) -284.99400 (11) -285.00100 (and) -284.99700 (25) -285.00100 (r) 37.01830 (e) 39.98840 (gions) -284.98200 (in) -285 (an) -285.00100 (ima) 10.01300 (g) 10.00320 (e) 9.99343 (\054) ] TJ T* [ (and) -243.98300 (obtain) -244.00300 (the) -243.99100 (best) -243.98500 (detection) -244.01400 (r) 37.01960 (esults) -244.01100 (among) -244.01600 (systems) -244.01800 (that) -244.01300 (do) ] TJ T* [ (not) -250.02000 (use) -249.99100 (object) -249.98200 (pr) 44.98390 (oposals) -249.99700 (for) -250.01800 (object) -249.98200 (localization\056) ] TJ /R17 11.95520 Tf 35.90590 TL T* [ (1\056) -249.99000 (Intr) 18.01460 (oduction) ] TJ /R19 9.96260 Tf 11.95510 -18.94300 Td [ (The) -354.01500 (process) -352.99000 (of) -354.01800 (localizing) -353.98900 (objects) -353.01700 (with) -353.99700 (bounding) -354.01700 (box) 14.99260 (es) ] TJ -11.95510 -11.95510 Td [ (can) -222.00600 (be) -222.00200 (seen) -222 (as) -222 (a) -222.00700 (control) -222.01900 (problem) -222.01400 (with) -222.01900 (a) -222.00700 (sequence) -221.99700 (of) -222 (steps) -221.98300 (to) ] TJ 11.95510 TL T* [ <7265026e65> -243.99100 (the) -243.99100 (geometry) -244.99600 (of) -243.99700 (the) -243.99100 (box\056) -308.01000 (Determining) -243.99900 (the) -243.99100 (e) 15.01220 (xact) -245.01500 (loca\055) ] TJ T* [ (tion) -234.01500 (of) -233.99800 (a) -234.00400 (tar) 17.99640 (get) -233.01300 (object) -233.98400 (in) -233.98800 (a) -234.00500 (scene) -234.00300 (requires) -233.99300 (acti) 24.98110 (v) 14.98280 (e) -234.00500 (eng) 4.98446 (agement) ] TJ T* [ (to) -199.99400 (understand) -200.01800 (the) -199.99800 (conte) 14.98650 (xt\054) -209.98900 (change) -200.00300 (the) -199.99800 <02786174696f6e> -200.00600 (point\054) -210.01200 (identify) ] TJ 11.95590 TL T* [ (distincti) 24.98480 (v) 14.98280 (e) -238.98400 (parts) -237.98800 (that) -239.00400 (support) -238.98800 (recognit) 0.99738 (ion\054) -240.99900 (and) -239.01400 (determine) -239 (the) ] TJ 11.95510 TL T* [ (correct) -250.00500 (proportions) -249.99300 (of) -249.99600 (the) -249.99000 (box\056) ] TJ 11.95510 -11.96800 Td [ (During) -316.00400 (the) -316.01800 (last) -316.01000 (decade\054) -333.00800 (the) -316.01800 (problem) -315.99900 (of) -317.00300 (object) -316.01100 (detection) ] TJ -11.95510 -11.95630 Td [ (or) -360.01700 (localization) -359.98500 (has) -360.99100 (been) -360.01700 (studied) -359.98800 (by) -360.00900 (the) -360.99100 (vis) 0.99738 (ion) -360.98100 (community) ] TJ T* [ (with) -286.00900 (the) -287.00300 (goal) -286.01800 (of) -285.99000 (recognizing) -286.98800 (the) -285.98400 (cate) 15.01100 (gory) -286.99900 (of) -285.98900 (an) -285.99100 (object\054) -296.01200 (and) ] TJ T* [ (identifying) -277.98600 (its) -278.99100 (spatial) -278.00800 (e) 15.01220 (xtent) -278.99300 (with) -278.01000 (a) -277.99800 (tight) -278.98000 (bounding) -277.99000 (box) -278.99800 (that) ] TJ T* [ (co) 15.01770 (v) 14.98280 (ers) -357.98100 (all) -358.00900 (its) -357.99800 (visible) -357.98600 (parts) -358.00800 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 162.28100 188.61000 Tm (10) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 172.24300 188.61000 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 178.30100 188.61000 Tm (26) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 188.26300 188.61000 Tm [ (\135\056) -634.99300 (This) -357.99700 (is) -358.00600 (a) -357.98400 (challenging) ] TJ -138.15100 -11.95510 Td [ (setup) -213.98500 (that) -215.00800 (requires) -213.99500 (computation) -213.99200 (and) -215.01700 (analysis) -214.01600 (in) -213.99100 (multiple) -215.01600 (im\055) ] TJ 11.95510 TL T* [ (age) -283.99600 (re) 15.00980 (gions\054) -293.01800 (and) -283.98600 (a) -283.99700 (good) -284.99200 (e) 15.01220 (xample) -284.01000 (of) -283.99100 (a) -283.99700 (task) -284.01600 (dri) 24.98600 (v) 14.98280 (en) -283.99200 (by) -285.00100 (acti) 24.98110 (v) 14.98280 (e) ] TJ 11.95590 TL (attention\056) ' 11.95510 -11.96800 Td [ (Important) -420.01700 (progress) -419.01200 (for) -420.01000 (impro) 15.00360 (ving) -419.98600 (the) -418.98200 (accurac) 14.99750 (y) -420.00300 (of) -420.00600 (ob\055) ] TJ -11.95510 -11.95510 Td [ (ject) -213.99800 (detectors) -213.98900 (has) -213.99600 (been) -214.00200 (recently) -215.01400 (poss) 1.00228 (ible) -215.00800 (with) -213.98100 (Con) 40.01540 (v) 20.00160 (olutional) ] TJ T* [ (Neural) -364.01500 (Netw) 10.00690 (orks) -362.98600 (\050CNNs\051\054) -393.01200 (which) -362.98200 (le) 25.01790 (v) 14.98280 (erage) -363.99100 (big) -364 (visual) -363.99100 (data) ] TJ 11.95510 TL T* [ (and) -356.01400 (deep) -356.01800 (learning) -356.00200 (for) -355 (image) -356.00300 (cate) 15.01100 (gorization\056) -627.98100 (A) -356.01700 (successful) ] TJ T* [ (model) -353.99400 (is) -354.98600 (the) -354.01100 (R\055CNN) -353.99000 (detector) -354.01100 (proposed) -355.01400 (by) -354.01000 (Girshick) -353.99500 (et) -354.99700 (al\056) ] TJ (\133) ' ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 53.42970 81 Tm (12) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 63.39220 81 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 69.43950 81 Tm (25) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 79.40200 81 Tm [ (\135\054) -384.01500 (which) -358.00300 (combines) -357.00700 (object) -356.98300 (proposals) -357.00700 (and) -358.01400 (CNN) -357.01900 (fea\055) ] TJ ET Q q 3088.62000 4804.85000 2362.46000 837.96500 re W* n 1 g 2543.44000 3381.31000 3634.56000 2725.92000 re f* q 373.83500 0 0 509.14100 3353.08000 5026.20000 cm /R25 Do Q q 739.28000 0 0 509.14100 3842.34000 5018.83000 cm /R26 Do Q q 746.24600 0 0 509.14100 4699.79000 5018.83000 cm /R27 Do Q 0.29004 0.49390 0.73242 rg 3421.79000 4961.09000 m 5425.79000 4961.09000 l 5425.79000 4957.30000 l 3421.79000 4957.30000 l h 5396.36000 4978.50000 m 5429.48000 4959.22000 l 5396.36000 4939.88000 l 5395.46000 4939.38000 5394.29000 4939.68000 5393.79000 4940.59000 c 5393.29000 4941.50000 5393.59000 4942.61000 5394.45000 4943.16000 c 5424.73000 4960.83000 l 5424.73000 4957.55000 l 5394.45000 4975.22000 l 5393.59000 4975.77000 5393.29000 4976.94000 5393.79000 4977.79000 c 5394.29000 4978.70000 5395.46000 4979.01000 5396.36000 4978.50000 c f 3.78600 w 1 j 0.29004 0.49390 0.73242 RG 3536.58000 4989.30000 m 3536.58000 4911.57000 l S 4021.59000 4989.30000 m 4021.59000 4911.57000 l S 4872.48000 4989.30000 m 4872.48000 4911.57000 l S 0 g q 10 0 0 10 0 0 cm BT /R29 9.09852 Tf 1 0 0 1 349.90700 482.49400 Tm <0750> Tj /R29 6.67548 Tf 3.27109 -1.87773 Td <0b35> Tj /R32 9.09852 Tf 4.17969 1.87773 Td ( ) Tj /R29 9.09852 Tf 41.59020 0 Td <0750> Tj /R29 6.67548 Tf 3.45313 -1.87773 Td <0bdc> Tj /R32 9.09852 Tf 2.96797 1.87773 Td ( ) Tj /R29 9.09852 Tf 69.78360 0 Td <0750> Tj /R29 6.67548 Tf 3.45313 -1.87773 Td <0be1> Tj /R29 9.09852 Tf 6.84492 1.87773 Td [ <0d46> -225.04300 <0373> ] TJ /R32 9.09852 Tf 13.87190 0 Td ( ) Tj ET Q 4393.63000 4989.30000 m 4393.63000 4911.57000 l S q 10 0 0 10 0 0 cm BT /R29 9.08640 Tf 1 0 0 1 428.25700 482.76200 Tm <0750> Tj /R29 6.66336 Tf 3.45273 -1.88789 Td <0bdc> Tj /R29 9.08640 Tf 4.96719 1.88789 Td [ <0d45> -219.67400 <0373> ] TJ /R32 9.08640 Tf 13.87190 0 Td ( ) Tj ET Q 5245.78000 4989.30000 m 5245.78000 4911.57000 l S q 10 0 0 10 0 0 cm BT /R29 9.08640 Tf 1 0 0 1 520.46400 482.79700 Tm <0750> Tj /R29 6.66336 Tf 3.45273 -1.87773 Td <0be1> Tj /R32 9.08640 Tf 4.78555 1.87773 Td ( ) Tj /R29 16.17380 Tf -157.78500 38.76870 Td <01e5> Tj /R32 16.17380 Tf 12.17580 0 Td ( ) Tj /R29 16.17380 Tf 73.79140 0.06055 Td <01e5> Tj /R32 16.17380 Tf 12.17580 0 Td ( ) Tj /R29 16.17380 Tf -98.01170 -39.45510 Td <01e5> Tj /R32 16.17380 Tf 12.19060 0 Td ( ) Tj /R29 16.18590 Tf 73.79180 0.06055 Td <01e5> Tj /R32 16.18590 Tf 12.17580 0 Td ( ) Tj /R32 9.08640 Tf -160.33400 24.27580 Td [ (Act) 4.00882 (i) 2.01199 (ons) -3.65953 ( ) ] TJ /R32 9.09852 Tf 2.50859 24.23050 Td [ (St) 14.98360 (a) 6.00791 (t) 14.97820 (es) -2.37204 ( ) ] TJ 1.22695 -47.90040 Td [ (St) 14.98360 (ep) 3.98739 (s) -2.47937 ( ) ] TJ /R32 9.08640 Tf 31.70120 75.17460 Td [ (Seq) -3.99539 (ue) -3.00124 (nce ) -7.02082 (of ) -7.98272 (a) 5.98099 (t) 14.98200 (t) 14.98200 (en) -3.00393 (de) -2.99856 (d r) 12.99370 (eg) -3.00393 (i) 2.00979 (ons) -3.01468 ( ) -6.01861 (t) 8.01765 (o l) 1.99366 (oc) 9.97907 (al) -1 (i) 4.98147 (z) 22.01630 (e ) -7.99078 (the ) -8.00690 (obje) -2.99856 (c) 2.98781 (t) -14.38020 ( ) ] TJ ET Q Q q 10 0 0 10 0 0 cm BT /R19 8.96640 Tf 1 0 0 1 308.86200 459.56300 Tm [ (Figure) -299.99700 (1\056) -300.01300 (A) -300.00200 (sequence) -301.02100 (of) -300.00500 (actions) -300.00800 (tak) 10.00920 (en) -300.01300 (by) -299.99100 (the) -300.00200 (proposed) -300.02400 (algorithm) ] TJ 10.95780 TL T* [ (to) -274.01800 (localize) -272.98600 (a) -274.00200 (co) 24.98750 (w) 64.98870 (\056) -381.00100 (The) -274.00700 (algorithm) -273.98300 (attends) -272.99700 (re) 14.99740 (gions) -274.01600 (and) -274.00500 (decides) -272.98100 (ho) 25.00660 (w) ] TJ 10.95900 TL T* [ (to) -226.00900 (transform) -225.01300 (the) -225.98500 (bounding) -225.01000 (box) -226.01700 (to) -226.00700 (progressi) 25.01740 (v) 14.98650 (ely) -224.98300 (localize) -225.97900 (the) -224.98300 (object\056) ] TJ /R19 9.96260 Tf 32.84300 TL T* [ (tures) -221.99200 (to) -221.99000 (achie) 25.01540 (v) 14.98280 (e) -222.00700 (state\055of\055the\055art) -221.01200 (results) -222.01700 (in) -221.99200 (the) -221.99200 (P) 14.99260 (ascal) -222.00200 (and) -221.99700 (Im\055) ] TJ 11.95510 TL T* [ (ageNet) -408.98800 (benchmarks\056) -786.98100 (Se) 24.98600 (v) 14.98280 (eral) -408.99500 (other) -408.98100 (w) 10.00320 (orks) -409.99500 (ha) 19.99670 (v) 14.98280 (e) -408.99500 (proposed) ] TJ T* [ (the) -366.99000 (use) -368.01000 (of) -366.99500 (high\055capacity) -367.99000 (CNNs) -367.01200 (to) -368.00700 (directly) -367.01700 (predict) -368.00200 (bound\055) ] TJ T* [ (ing) -292.99300 (box) 14.99260 (es) -293.00700 (under) -291.98300 (a) -293.01500 (re) 15.00980 (gression) -292.98500 (setting) -291.99500 (also) -292.99500 (with) -292.98500 (good) -292.99000 (results) ] TJ 11.95590 TL (\133) ' ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 312.18000 356.98200 Tm (27) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 322.14200 356.98200 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 327.12300 356.98200 Tm (28) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 337.08600 356.98200 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 342.06700 356.98200 Tm (9) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 347.04800 356.98200 Tm (\135\056) Tj -26.23120 -12.46210 Td [ (In) -376.01300 (this) -377.01100 (w) 10 (ork\054) -406.98100 (we) -376.99800 (propose) -375.99600 (a) -375.98100 <636c6173732d73706563690263> -375.98600 (acti) 24.98110 (v) 14.98280 (e) -377 (detec\055) ] TJ -11.95510 -11.95470 Td [ (tion) -331.01800 (model) -331.01800 (that) ] TJ /R23 9.96260 Tf 65.25080 0 Td (learns) Tj /R19 9.96260 Tf 28.20630 0 Td [ (to) -330.99400 (localize) -330.99400 (tar) 17.99700 (get) -332.01300 (object) 0.99493 (s) -332.01800 (kno) 24.99090 (wn) -330.99400 (by) ] TJ -93.45700 -11.95510 Td [ (the) -237.01200 (system\056) -305.01300 (The) -237.01400 (proposed) -236.01500 (model) -236.99500 (follo) 24.99580 (ws) -235.99000 (a) -236.98500 (top\055do) 25 (wn) -237.00900 (search) ] TJ 11.95510 TL T* [ (strate) 14.98520 (gy) 65.01860 (\054) -216.98100 (which) -207.99000 (starts) -207.98500 (by) -207.99500 (analyzing) -207.98500 (the) -207.99500 (whole) -207.99000 (scene) -208.98500 (and) -208 (then) ] TJ T* [ (proceeds) -404.98600 (to) -405.01800 (narro) 25.01540 (w) -406.00800 (do) 24.98600 (wn) -404.98100 (the) -404.98100 (correct) -404.99600 (location) -405.00100 (of) -406.01100 (objects\056) ] TJ T* [ (This) -254.01400 (is) -255.00400 (achie) 25.01540 (v) 14.98280 (ed) -253.99700 (by) -255.00600 (applying) -254.00200 (a) -254.98200 (sequence) -253.99200 (of) -255.01600 (transformations) ] TJ 11.95590 TL T* [ (to) -241.00600 (a) -240.98400 (box) -241.00400 (that) -241.00400 (initially) -241.00900 (co) 15.01710 (v) 14.98280 (ers) -240.97900 (a) -240.98400 (lar) 17.99700 (ge) -241.01900 (re) 15.00730 (gion) -240.99400 (of) -241.01900 (the) -241.00900 (image) -241.00400 (and) ] TJ 11.95510 TL T* [ (is) -360.00600 <026e616c6c79> -359.98400 (reduced) -360.02100 (to) -360.98600 (a) -359.98400 (tight) -359.98900 (bounding) -360.01300 (box\056) -641.01100 (The) -360.01300 (sequence) ] TJ T* [ (of) -277.99000 (transform) 1.00228 (ations) -278.00500 (is) -278.02000 (decided) -277.00300 (by) -277.98300 (an) ] TJ /R23 9.96260 Tf 144.41900 0 Td [ (a) 10.00320 (g) 10.00320 (ent) ] TJ /R19 9.96260 Tf 24.88200 0 Td [ (that) -278.01700 (analyzes) -277.00800 (the) ] TJ -169.30100 -11.95510 Td [ (content) -230.02000 (of) -231.01800 (the) -229.99300 (currently) -229.98600 (visible) -230.00600 (re) 15.00730 (gion) -230.99600 (to) -229.99100 (select) -229.99100 (the) -229.99100 (ne) 15.01710 (xt) -231.01000 (best) ] TJ T* [ (action\056) -550.98900 (Each) -329.98400 (transformation) -329.98400 (should) -331.00800 (k) 10.00320 (eep) -329.98900 (the) -330.01400 (object) -330.98900 (inside) ] TJ T* [ (the) -281.98500 (visible) -281.01700 (re) 15.00980 (gion) -282.00900 (while) -282.01400 (cutt) 0.99493 (ing) -282.01200 (of) 25.01540 (f) -281.99700 (as) -281.01200 (much) -282.01200 (background) -282.00700 (as) ] TJ 11.95590 TL T* [ (possible\056) -309.98700 (Figure) ] TJ ET Q 1 0 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 375.58200 201.05800 Tm (1) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 383.04400 201.05800 Tm [ (illustrates) -247.99800 (some) -248.99800 (steps) -248.99800 (of) -247.99800 (the) -249.00700 (dynamic) -248.99300 (de\055) ] TJ -74.18160 -11.95510 Td [ (cision) -250.01000 (process) -249.98800 (to) -249.98500 (localize) -249.98800 (a) -250.00200 (co) 25.01540 (w) -249.99300 (in) -249.98800 (an) -249.99700 (image\056) ] TJ 11.95510 -12.46210 Td [ (The) -303.98400 (proposed) -302.98400 (approach) -303.98400 (is) -304.01300 (fundament) 1.00964 (ally) -304.01300 (dif) 24.98600 (ferent) -303.99300 (from) ] TJ -11.95510 -11.95510 Td [ (most) -240.98900 (localization) -242.00400 (str) 0.99003 (ate) 15.01470 (gies\056) -308.01200 (In) -241.01900 (contrast) -240.99900 (to) -241.00900 (sliding) -241.99400 (windo) 25 (ws\054) ] TJ 11.95470 TL T* [ (our) -241.99100 (approach) -243.01300 (does) -241.98700 (not) -243.00100 (follo) 24.99580 (w) -241.99400 (a) -242.98400 <0278> 14.99750 (ed) -241.99900 (path) -243.00400 (to) -241.98900 (search) -242.98400 (objects\073) ] TJ T* [ (instead\054) -324 (dif) 24.98600 (ferent) -309.01000 (objects) -308.98500 (in) -308.99700 (dif) 24.98600 (ferent) -309.01200 (scenes) -309.00200 (will) -308.98300 (end) -309.00200 (up) -308.99700 (in) ] TJ T* [ (dif) 24.98600 (ferent) -337.00500 (search) -335.98800 (paths\056) -570 (Unlik) 9.98853 (e) -337.00700 (object) -336.00800 (proposal) -337.01700 (algorithms\054) ] TJ 11.95590 TL T* [ (candidate) -352.98800 (re) 15.00980 (gions) -353.99000 (in) -352.98800 (our) -352.99300 (approach) -354.01400 (a) 1.01454 (re) -353.99000 (selected) -352.99000 (by) -352.99000 (a) -353.98500 (high\055) ] TJ 11.95510 TL T* [ (le) 25.01790 (v) 14.98280 (el) -214.98100 (reasoning) -213.98100 (strate) 14.98520 (gy) -215.01300 (instead) -213.98400 (of) -214.98400 (follo) 24.99580 (wing) -213.98400 (lo) 24.98600 (w\055le) 25.01540 (v) 14.98280 (el) -214.97900 (cues\056) ] TJ T* [ (Also\054) -234.01500 (compared) -229.99100 (to) -231.00800 (bounding) -229.99600 (box) -229.98600 (re) 15.00730 (gression) -230.99600 (al) 0.98513 (gorithms\054) -235.01000 (our) ] TJ T* [ (approach) -221.01700 (does) -219.99000 (not) -221.00500 (localize) -219.99300 (objects) -221.00200 (follo) 24.99580 (wing) -221.00200 (a) -220.00700 (single\054) -226.99200 (struc\055) ] 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 (2488) 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 /DCTDecode /Interpolate true /Height 200 /Length 10475 /ColorSpace /DeviceRGB /Width 293 /BitsPerComponent 8 /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 ? *)_˦ccЀ~U2Jj;>wj