%PDF-1.3
1 0 obj
<<
/Kids [ 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 13 0 R 14 0 R ]
/Type /Pages
/Count 11
>>
endobj
2 0 obj
<<
/Subject (Neural Information Processing Systems http\072\057\057nips\056cc\057)
/Publisher (Curran Associates\054 Inc\056)
/Language (en\055US)
/Created (2018)
/EventType (Poster)
/Description-Abstract (While the use of bottom\055up local operators in convolutional neural networks \050CNNs\051 matches well some of the statistics of natural images\054 it may also prevent such models from capturing contextual long\055range feature interactions\056 In this work\054 we propose a simple\054 lightweight approach for better context exploitation in CNNs\056 We do so by introducing a pair of operators\072 gather\054 which efficiently aggregates feature responses from a large spatial extent\054 and excite\054 which redistributes the pooled information to local features\056 The operators are cheap\054 both in terms of number of added parameters and computational complexity\054 and can be integrated directly in existing architectures to improve their performance\056 Experiments on several datasets show that gather\055excite can bring benefits comparable to increasing the depth of a CNN at a fraction of the cost\056 For example\054 we find ResNet\05550 with gather\055excite operators is able to outperform its 101\055layer counterpart on ImageNet with no additional learnable parameters\056 We also propose a parametric gather\055excite operator pair which yields further performance gains\054 relate it to the recently\055introduced Squeeze\055and\055Excitation Networks\054 and analyse the effects of these changes to the CNN feature activation statistics\056)
/Producer (PyPDF2)
/Title (Gather\055Excite\072 Exploiting Feature Context in Convolutional Neural Networks)
/Date (2018)
/ModDate (D\07220190219012858\05508\04700\047)
/Published (2018)
/Type (Conference Proceedings)
/firstpage (9401)
/Book (Advances in Neural Information Processing Systems 31)
/Description (Paper accepted and presented at the Neural Information Processing Systems Conference \050http\072\057\057nips\056cc\057\051)
/Editors (S\056 Bengio and H\056 Wallach and H\056 Larochelle and K\056 Grauman and N\056 Cesa\055Bianchi and R\056 Garnett)
/Author (Jie Hu\054 Li Shen\054 Samuel Albanie\054 Gang Sun\054 Andrea Vedaldi)
/lastpage (9411)
>>
endobj
3 0 obj
<<
/Type /Catalog
/Pages 1 0 R
>>
endobj
4 0 obj
<<
/Contents 15 0 R
/Parent 1 0 R
/Resources 16 0 R
/MediaBox [ 0 0 612 792 ]
/Annots [ 47 0 R 48 0 R 49 0 R 50 0 R 51 0 R 52 0 R 53 0 R 54 0 R 55 0 R 56 0 R 57 0 R 58 0 R ]
/Type /Page
>>
endobj
5 0 obj
<<
/Contents 59 0 R
/Parent 1 0 R
/Resources 60 0 R
/MediaBox [ 0 0 612 792 ]
/Annots [ 130 0 R 131 0 R 132 0 R 133 0 R 134 0 R 135 0 R 136 0 R ]
/Type /Page
>>
endobj
6 0 obj
<<
/Contents 137 0 R
/Parent 1 0 R
/Resources 138 0 R
/MediaBox [ 0 0 612 792 ]
/Annots [ 199 0 R 200 0 R 201 0 R 202 0 R 203 0 R ]
/Type /Page
>>
endobj
7 0 obj
<<
/Contents 204 0 R
/Parent 1 0 R
/Resources 205 0 R
/MediaBox [ 0 0 612 792 ]
/Annots [ 214 0 R 215 0 R 216 0 R 217 0 R 218 0 R 219 0 R 220 0 R 221 0 R ]
/Type /Page
>>
endobj
8 0 obj
<<
/Contents 222 0 R
/Parent 1 0 R
/Resources 223 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 ]
/Type /Page
>>
endobj
9 0 obj
<<
/Contents 240 0 R
/Parent 1 0 R
/Resources 241 0 R
/MediaBox [ 0 0 612 792 ]
/Annots [ 246 0 R 247 0 R 248 0 R 249 0 R 250 0 R 251 0 R 252 0 R 253 0 R 254 0 R 255 0 R 256 0 R 257 0 R 258 0 R 259 0 R 260 0 R 261 0 R 262 0 R ]
/Type /Page
>>
endobj
10 0 obj
<<
/Contents 263 0 R
/Parent 1 0 R
/Resources 264 0 R
/MediaBox [ 0 0 612 792 ]
/Annots [ 367 0 R 368 0 R 369 0 R 370 0 R 371 0 R 372 0 R 373 0 R ]
/Type /Page
>>
endobj
11 0 obj
<<
/Contents 374 0 R
/Parent 1 0 R
/Resources 375 0 R
/MediaBox [ 0 0 612 792 ]
/Annots [ 424 0 R 425 0 R 426 0 R 427 0 R 428 0 R 429 0 R 430 0 R 431 0 R 432 0 R 433 0 R 434 0 R 435 0 R 436 0 R 437 0 R 438 0 R 439 0 R 440 0 R 441 0 R 442 0 R ]
/Type /Page
>>
endobj
12 0 obj
<<
/Contents 443 0 R
/Parent 1 0 R
/Resources 444 0 R
/MediaBox [ 0 0 612 792 ]
/Annots [ 445 0 R 446 0 R 447 0 R 448 0 R 449 0 R 450 0 R 451 0 R 452 0 R 453 0 R 454 0 R 455 0 R 456 0 R 457 0 R 458 0 R 459 0 R 460 0 R 461 0 R 462 0 R 463 0 R 464 0 R 465 0 R ]
/Type /Page
>>
endobj
13 0 obj
<<
/Contents 466 0 R
/Parent 1 0 R
/Resources 467 0 R
/MediaBox [ 0 0 612 792 ]
/Annots [ 468 0 R 469 0 R 470 0 R 471 0 R 472 0 R 473 0 R 474 0 R 475 0 R 476 0 R 477 0 R 478 0 R 479 0 R 480 0 R 481 0 R 482 0 R 483 0 R 484 0 R 485 0 R 486 0 R 487 0 R 488 0 R 489 0 R 490 0 R 491 0 R 492 0 R 493 0 R 494 0 R 495 0 R 496 0 R 497 0 R 498 0 R 499 0 R 500 0 R 501 0 R 502 0 R ]
/Type /Page
>>
endobj
14 0 obj
<<
/Contents 503 0 R
/Parent 1 0 R
/Resources 504 0 R
/MediaBox [ 0 0 612 792 ]
/Annots [ 505 0 R 506 0 R 507 0 R 508 0 R 509 0 R 510 0 R 511 0 R 512 0 R 513 0 R 514 0 R 515 0 R 516 0 R 517 0 R 518 0 R 519 0 R 520 0 R 521 0 R 522 0 R 523 0 R 524 0 R 525 0 R 526 0 R 527 0 R 528 0 R 529 0 R 530 0 R 531 0 R ]
/Type /Page
>>
endobj
15 0 obj
<<
/Length 3301
/Filter /FlateDecode
>>
stream
xZK6ϯБ!dSn6&PBRO~~)N"/
7MyyH<Íl#Mq!Β7`i/>fݩ8۪8p߶`?<6q(Dmv2m0xm}k^=wѡt͋I&7%&~"W$n@J'X
0q0y?TtWI ,ՆҊ!Dzvކf$Z둈Uĸlro2GWJAd~}%"!"PPQ1̢KH"Y26n
ݓuznTd4:,imW^Elɺ]{h?@&DdT^:?>oeH@VF4]*gLX;";eL/7¿Y)SJ%zn83nbkd'3g-i4)\U{{*(*/檒r[D:(/~bps4V"T;0IL~=T5|6i/ @wmARd|hna1l ߗ
XH}?)!(PlO<ںfp2i Ǟꇪ
:g V|o['eЩhJ(