id sid tid token lemma pos fcis-1460 1 1 frontiers frontier NOUN fcis-1460 1 2 in in ADP fcis-1460 1 3 computing computing NOUN fcis-1460 1 4 and and CCONJ fcis-1460 1 5 intelligent intelligent ADJ fcis-1460 1 6 systems system NOUN fcis-1460 1 7 issn issn VERB fcis-1460 1 8 : : PUNCT fcis-1460 1 9 2832 2832 NUM fcis-1460 1 10 - - SYM fcis-1460 1 11 6024 6024 NUM fcis-1460 1 12 | | NOUN fcis-1460 1 13 vol vol NOUN fcis-1460 1 14 . . PROPN fcis-1460 2 1 1 1 NUM fcis-1460 2 2 , , PUNCT fcis-1460 2 3 no no INTJ fcis-1460 2 4 . . NOUN fcis-1460 2 5 1 1 NUM fcis-1460 2 6 , , PUNCT fcis-1460 2 7 2022 2022 NUM fcis-1460 2 8 45 45 NUM fcis-1460 2 9 discnet discnet NOUN fcis-1460 2 10 : : PUNCT fcis-1460 2 11 pixel pixel PROPN fcis-1460 2 12 segmentation segmentation NOUN fcis-1460 2 13 based base VERB fcis-1460 2 14 on on ADP fcis-1460 2 15 discrete discrete ADJ fcis-1460 2 16 features feature NOUN fcis-1460 2 17 zhihui zhihui PROPN fcis-1460 2 18 li1 li1 NOUN fcis-1460 2 19 , , PUNCT fcis-1460 2 20 * * PUNCT fcis-1460 2 21 , , PUNCT fcis-1460 2 22 xiaoshuo xiaoshuo PROPN fcis-1460 2 23 jia2 jia2 PROPN fcis-1460 2 24 1 1 NUM fcis-1460 2 25 school school NOUN fcis-1460 2 26 of of ADP fcis-1460 2 27 computer computer NOUN fcis-1460 2 28 science science NOUN fcis-1460 2 29 , , PUNCT fcis-1460 2 30 guangdong guangdong PROPN fcis-1460 2 31 university university PROPN fcis-1460 2 32 of of ADP fcis-1460 2 33 science science NOUN fcis-1460 2 34 and and CCONJ fcis-1460 2 35 technology technology NOUN fcis-1460 2 36 , , PUNCT fcis-1460 2 37 dongguan dongguan PROPN fcis-1460 2 38 523079 523079 NUM fcis-1460 2 39 , , PUNCT fcis-1460 2 40 guangdong guangdong PROPN fcis-1460 2 41 , , PUNCT fcis-1460 2 42 china china PROPN fcis-1460 2 43 2 2 NUM fcis-1460 2 44 electrical electrical ADJ fcis-1460 2 45 engineering engineering NOUN fcis-1460 2 46 department department NOUN fcis-1460 2 47 , , PUNCT fcis-1460 2 48 university university PROPN fcis-1460 2 49 of of ADP fcis-1460 2 50 colorado colorado PROPN fcis-1460 2 51 , , PUNCT fcis-1460 2 52 boulder boulder NOUN fcis-1460 2 53 , , PUNCT fcis-1460 2 54 co co NOUN fcis-1460 2 55 80309 80309 NUM fcis-1460 2 56 , , PUNCT fcis-1460 2 57 usa usa PROPN fcis-1460 2 58 * * PUNCT fcis-1460 2 59 corresponding correspond VERB fcis-1460 2 60 author author NOUN fcis-1460 2 61 : : PUNCT fcis-1460 2 62 zhihui zhihui PROPN fcis-1460 2 63 li li PROPN fcis-1460 2 64 . . PROPN fcis-1460 3 1 abstract abstract PROPN fcis-1460 3 2 . . PUNCT fcis-1460 4 1 in in ADP fcis-1460 4 2 the the DET fcis-1460 4 3 process process NOUN fcis-1460 4 4 of of ADP fcis-1460 4 5 pixel pixel ADJ fcis-1460 4 6 - - PUNCT fcis-1460 4 7 level level NOUN fcis-1460 4 8 semantic semantic ADJ fcis-1460 4 9 segmentation segmentation NOUN fcis-1460 4 10 tasks task NOUN fcis-1460 4 11 , , PUNCT fcis-1460 4 12 traditional traditional ADJ fcis-1460 4 13 image image NOUN fcis-1460 4 14 processing processing NOUN fcis-1460 4 15 algorithms algorithm NOUN fcis-1460 4 16 will will AUX fcis-1460 4 17 suffer suffer VERB fcis-1460 4 18 from from ADP fcis-1460 4 19 the the DET fcis-1460 4 20 working work VERB fcis-1460 4 21 mechanism mechanism NOUN fcis-1460 4 22 of of ADP fcis-1460 4 23 convolutional convolutional ADJ fcis-1460 4 24 layers layer NOUN fcis-1460 4 25 and and CCONJ fcis-1460 4 26 pooling pool VERB fcis-1460 4 27 layers layer NOUN fcis-1460 4 28 , , PUNCT fcis-1460 4 29 resulting result VERB fcis-1460 4 30 in in ADP fcis-1460 4 31 the the DET fcis-1460 4 32 loss loss NOUN fcis-1460 4 33 of of ADP fcis-1460 4 34 some some DET fcis-1460 4 35 features feature NOUN fcis-1460 4 36 , , PUNCT fcis-1460 4 37 which which PRON fcis-1460 4 38 now now ADV fcis-1460 4 39 leads lead VERB fcis-1460 4 40 to to PART fcis-1460 4 41 inaccurate inaccurate VERB fcis-1460 4 42 semantic semantic ADJ fcis-1460 4 43 segmentation segmentation NOUN fcis-1460 4 44 accuracy accuracy NOUN fcis-1460 4 45 . . PUNCT fcis-1460 5 1 for for ADP fcis-1460 5 2 such such ADJ fcis-1460 5 3 problems problem NOUN fcis-1460 5 4 , , PUNCT fcis-1460 5 5 we we PRON fcis-1460 5 6 design design VERB fcis-1460 5 7 a a DET fcis-1460 5 8 discrete discrete ADJ fcis-1460 5 9 pooling pooling NOUN fcis-1460 5 10 layer layer NOUN fcis-1460 5 11 by by ADP fcis-1460 5 12 analyzing analyze VERB fcis-1460 5 13 the the DET fcis-1460 5 14 distribution distribution NOUN fcis-1460 5 15 and and CCONJ fcis-1460 5 16 statistical statistical ADJ fcis-1460 5 17 properties property NOUN fcis-1460 5 18 of of ADP fcis-1460 5 19 discrete discrete ADJ fcis-1460 5 20 data datum NOUN fcis-1460 5 21 . . PUNCT fcis-1460 6 1 compared compare VERB fcis-1460 6 2 with with ADP fcis-1460 6 3 the the DET fcis-1460 6 4 traditional traditional ADJ fcis-1460 6 5 pooling pool VERB fcis-1460 6 6 layer layer NOUN fcis-1460 6 7 , , PUNCT fcis-1460 6 8 the the DET fcis-1460 6 9 discrete discrete ADJ fcis-1460 6 10 pooling pooling NOUN fcis-1460 6 11 layer layer NOUN fcis-1460 6 12 can can AUX fcis-1460 6 13 not not PART fcis-1460 6 14 only only ADV fcis-1460 6 15 preserve preserve VERB fcis-1460 6 16 the the DET fcis-1460 6 17 spatial spatial ADJ fcis-1460 6 18 information information NOUN fcis-1460 6 19 of of ADP fcis-1460 6 20 features feature NOUN fcis-1460 6 21 , , PUNCT fcis-1460 6 22 avoid avoid VERB fcis-1460 6 23 the the DET fcis-1460 6 24 loss loss NOUN fcis-1460 6 25 of of ADP fcis-1460 6 26 features feature NOUN fcis-1460 6 27 , , PUNCT fcis-1460 6 28 but but CCONJ fcis-1460 6 29 also also ADV fcis-1460 6 30 can can AUX fcis-1460 6 31 efficiently efficiently ADV fcis-1460 6 32 improve improve VERB fcis-1460 6 33 the the DET fcis-1460 6 34 accurate accurate ADJ fcis-1460 6 35 segmentation segmentation NOUN fcis-1460 6 36 of of ADP fcis-1460 6 37 instance instance NOUN fcis-1460 6 38 images image NOUN fcis-1460 6 39 . . PUNCT fcis-1460 7 1 then then ADV fcis-1460 7 2 , , PUNCT fcis-1460 7 3 based base VERB fcis-1460 7 4 on on ADP fcis-1460 7 5 the the DET fcis-1460 7 6 discrete discrete ADJ fcis-1460 7 7 pooling pool VERB fcis-1460 7 8 layer layer NOUN fcis-1460 7 9 , , PUNCT fcis-1460 7 10 we we PRON fcis-1460 7 11 design design VERB fcis-1460 7 12 discnet discnet NOUN fcis-1460 7 13 in in ADP fcis-1460 7 14 combination combination NOUN fcis-1460 7 15 with with ADP fcis-1460 7 16 the the DET fcis-1460 7 17 convolutional convolutional ADJ fcis-1460 7 18 layer layer NOUN fcis-1460 7 19 . . PUNCT fcis-1460 8 1 finally finally ADV fcis-1460 8 2 , , PUNCT fcis-1460 8 3 discnet discnet NOUN fcis-1460 8 4 is be AUX fcis-1460 8 5 compared compare VERB fcis-1460 8 6 with with ADP fcis-1460 8 7 some some DET fcis-1460 8 8 state state NOUN fcis-1460 8 9 - - PUNCT fcis-1460 8 10 of of ADP fcis-1460 8 11 - - PUNCT fcis-1460 8 12 the the DET fcis-1460 8 13 - - PUNCT fcis-1460 8 14 art art NOUN fcis-1460 8 15 algorithms algorithm NOUN fcis-1460 8 16 under under ADP fcis-1460 8 17 the the DET fcis-1460 8 18 cityscapes cityscape NOUN fcis-1460 8 19 dataset dataset VERB fcis-1460 8 20 . . PUNCT fcis-1460 9 1 experiments experiment NOUN fcis-1460 9 2 demonstrate demonstrate VERB fcis-1460 9 3 that that SCONJ fcis-1460 9 4 discnet discnet NOUN fcis-1460 9 5 achieves achieve VERB fcis-1460 9 6 excellent excellent ADJ fcis-1460 9 7 results result NOUN fcis-1460 9 8 in in ADP fcis-1460 9 9 both both DET fcis-1460 9 10 accuracy accuracy NOUN fcis-1460 9 11 and and CCONJ fcis-1460 9 12 speed speed NOUN fcis-1460 9 13 . . PUNCT fcis-1460 10 1 keywords keyword NOUN fcis-1460 10 2 : : PUNCT fcis-1460 10 3 semantic semantic ADJ fcis-1460 10 4 segmentation segmentation NOUN fcis-1460 10 5 , , PUNCT fcis-1460 10 6 image image NOUN fcis-1460 10 7 processing processing NOUN fcis-1460 10 8 , , PUNCT fcis-1460 10 9 pooling pool VERB fcis-1460 10 10 layer layer NOUN fcis-1460 10 11 , , PUNCT fcis-1460 10 12 discrete discrete ADJ fcis-1460 10 13 feature feature NOUN fcis-1460 10 14 . . PUNCT fcis-1460 11 1 1 1 X fcis-1460 11 2 . . X fcis-1460 11 3 introduction introduction NOUN fcis-1460 11 4 in in ADP fcis-1460 11 5 image image NOUN fcis-1460 11 6 processing processing NOUN fcis-1460 11 7 , , PUNCT fcis-1460 11 8 pixel pixel ADJ fcis-1460 11 9 - - PUNCT fcis-1460 11 10 level level NOUN fcis-1460 11 11 semantic semantic ADJ fcis-1460 11 12 segmentation segmentation NOUN fcis-1460 11 13 [ [ X fcis-1460 11 14 13 13 NUM fcis-1460 11 15 ] ] PUNCT fcis-1460 11 16 is be AUX fcis-1460 11 17 an an DET fcis-1460 11 18 extremely extremely ADV fcis-1460 11 19 important important ADJ fcis-1460 11 20 and and CCONJ fcis-1460 11 21 complex complex ADJ fcis-1460 11 22 task task NOUN fcis-1460 11 23 . . PUNCT fcis-1460 12 1 the the DET fcis-1460 12 2 cnn cnn PROPN fcis-1460 12 3 algorithm algorithm NOUN fcis-1460 12 4 [ [ X fcis-1460 12 5 4 4 NUM fcis-1460 12 6 - - SYM fcis-1460 12 7 7 7 NUM fcis-1460 12 8 ] ] PUNCT fcis-1460 12 9 has have AUX fcis-1460 12 10 achieved achieve VERB fcis-1460 12 11 excellent excellent ADJ fcis-1460 12 12 results result NOUN fcis-1460 12 13 in in ADP fcis-1460 12 14 image image NOUN fcis-1460 12 15 classification classification NOUN fcis-1460 12 16 , , PUNCT fcis-1460 12 17 segmentation segmentation NOUN fcis-1460 12 18 , , PUNCT fcis-1460 12 19 tracking tracking NOUN fcis-1460 12 20 and and CCONJ fcis-1460 12 21 other other ADJ fcis-1460 12 22 aspects aspect NOUN fcis-1460 12 23 of of ADP fcis-1460 12 24 kaggle kaggle NOUN fcis-1460 12 25 and and CCONJ fcis-1460 12 26 ai ai VERB fcis-1460 12 27 challenger challenger NOUN fcis-1460 12 28 competitions competition NOUN fcis-1460 12 29 using use VERB fcis-1460 12 30 the the DET fcis-1460 12 31 characteristics characteristic NOUN fcis-1460 12 32 of of ADP fcis-1460 12 33 multi multi NOUN fcis-1460 12 34 - - NOUN fcis-1460 12 35 parameters parameter NOUN fcis-1460 12 36 . . PUNCT fcis-1460 13 1 fcn fcn PROPN fcis-1460 13 2 uses use VERB fcis-1460 13 3 deconvolution deconvolution NOUN fcis-1460 13 4 for for ADP fcis-1460 13 5 upsampling upsample VERB fcis-1460 13 6 to to PART fcis-1460 13 7 make make VERB fcis-1460 13 8 the the DET fcis-1460 13 9 extracted extract VERB fcis-1460 13 10 features feature NOUN fcis-1460 13 11 more more ADV fcis-1460 13 12 detailed detailed ADJ fcis-1460 13 13 . . PUNCT fcis-1460 14 1 u u NOUN fcis-1460 14 2 - - NOUN fcis-1460 14 3 net net NOUN fcis-1460 14 4 uses use VERB fcis-1460 14 5 the the DET fcis-1460 14 6 network network NOUN fcis-1460 14 7 symmetric symmetric ADJ fcis-1460 14 8 structure structure NOUN fcis-1460 14 9 to to PART fcis-1460 14 10 fuse fuse VERB fcis-1460 14 11 highdimensional highdimensional NOUN fcis-1460 14 12 features feature NOUN fcis-1460 14 13 and and CCONJ fcis-1460 14 14 low low ADJ fcis-1460 14 15 - - PUNCT fcis-1460 14 16 dimensional dimensional ADJ fcis-1460 14 17 features feature NOUN fcis-1460 14 18 to to ADP fcis-1460 14 19 weight weight NOUN fcis-1460 14 20 edge edge NOUN fcis-1460 14 21 features feature NOUN fcis-1460 14 22 . . PUNCT fcis-1460 15 1 in in ADP fcis-1460 15 2 cpfnet cpfnet NOUN fcis-1460 15 3 , , PUNCT fcis-1460 15 4 the the DET fcis-1460 15 5 dilated dilated ADJ fcis-1460 15 6 convolution convolution NOUN fcis-1460 15 7 is be AUX fcis-1460 15 8 proposed propose VERB fcis-1460 15 9 , , PUNCT fcis-1460 15 10 which which PRON fcis-1460 15 11 can can AUX fcis-1460 15 12 expand expand VERB fcis-1460 15 13 the the DET fcis-1460 15 14 field field NOUN fcis-1460 15 15 of of ADP fcis-1460 15 16 view view NOUN fcis-1460 15 17 of of ADP fcis-1460 15 18 the the DET fcis-1460 15 19 convolutional convolutional ADJ fcis-1460 15 20 layer layer NOUN fcis-1460 15 21 to to PART fcis-1460 15 22 extract extract VERB fcis-1460 15 23 more more ADJ fcis-1460 15 24 feature feature NOUN fcis-1460 15 25 information information NOUN fcis-1460 15 26 , , PUNCT fcis-1460 15 27 and and CCONJ fcis-1460 15 28 then then ADV fcis-1460 15 29 combine combine VERB fcis-1460 15 30 the the DET fcis-1460 15 31 inception inception NOUN fcis-1460 15 32 module module NOUN fcis-1460 15 33 to to PART fcis-1460 15 34 achieve achieve VERB fcis-1460 15 35 context context NOUN fcis-1460 15 36 - - PUNCT fcis-1460 15 37 based base VERB fcis-1460 15 38 feature feature NOUN fcis-1460 15 39 fusion fusion NOUN fcis-1460 15 40 , , PUNCT fcis-1460 15 41 and and CCONJ fcis-1460 15 42 achieve achieve VERB fcis-1460 15 43 superior superior ADJ fcis-1460 15 44 results result NOUN fcis-1460 15 45 in in ADP fcis-1460 15 46 medical medical ADJ fcis-1460 15 47 datasets dataset NOUN fcis-1460 15 48 . . PUNCT fcis-1460 16 1 stdc stdc PROPN fcis-1460 16 2 is be AUX fcis-1460 16 3 based base VERB fcis-1460 16 4 on on ADP fcis-1460 16 5 fpn fpn VERB fcis-1460 16 6 for for ADP fcis-1460 16 7 the the DET fcis-1460 16 8 fusion fusion NOUN fcis-1460 16 9 of of ADP fcis-1460 16 10 multiple multiple ADJ fcis-1460 16 11 scales scale NOUN fcis-1460 16 12 , , PUNCT fcis-1460 16 13 so so SCONJ fcis-1460 16 14 its its PRON fcis-1460 16 15 performance performance NOUN fcis-1460 16 16 is be AUX fcis-1460 16 17 superior superior ADJ fcis-1460 16 18 to to ADP fcis-1460 16 19 the the DET fcis-1460 16 20 cpfnet cpfnet NOUN fcis-1460 16 21 algorithm algorithm NOUN fcis-1460 16 22 . . PUNCT fcis-1460 17 1 bisenetv2 bisenetv2 PROPN fcis-1460 17 2 adopts adopt VERB fcis-1460 17 3 a a DET fcis-1460 17 4 bilateral bilateral ADJ fcis-1460 17 5 segmentation segmentation NOUN fcis-1460 17 6 structure structure NOUN fcis-1460 17 7 on on ADP fcis-1460 17 8 the the DET fcis-1460 17 9 basis basis NOUN fcis-1460 17 10 of of ADP fcis-1460 17 11 stdc stdc NOUN fcis-1460 17 12 , , PUNCT fcis-1460 17 13 namely namely ADV fcis-1460 17 14 detail detail NOUN fcis-1460 17 15 branch branch NOUN fcis-1460 17 16 and and CCONJ fcis-1460 17 17 semantic semantic ADJ fcis-1460 17 18 branch branch NOUN fcis-1460 17 19 . . PUNCT fcis-1460 18 1 detail detail NOUN fcis-1460 18 2 branch branch NOUN fcis-1460 18 3 obtains obtain VERB fcis-1460 18 4 more more ADJ fcis-1460 18 5 low low ADJ fcis-1460 18 6 - - PUNCT fcis-1460 18 7 level level NOUN fcis-1460 18 8 feature feature NOUN fcis-1460 18 9 information information NOUN fcis-1460 18 10 by by ADP fcis-1460 18 11 expanding expand VERB fcis-1460 18 12 the the DET fcis-1460 18 13 channel channel NOUN fcis-1460 18 14 , , PUNCT fcis-1460 18 15 and and CCONJ fcis-1460 18 16 semantic semantic ADJ fcis-1460 18 17 branch branch NOUN fcis-1460 18 18 expands expand VERB fcis-1460 18 19 the the DET fcis-1460 18 20 receptive receptive ADJ fcis-1460 18 21 field field NOUN fcis-1460 18 22 through through ADP fcis-1460 18 23 a a DET fcis-1460 18 24 lightweight lightweight ADJ fcis-1460 18 25 convolution convolution NOUN fcis-1460 18 26 layer layer NOUN fcis-1460 18 27 to to PART fcis-1460 18 28 obtain obtain VERB fcis-1460 18 29 high high ADJ fcis-1460 18 30 - - PUNCT fcis-1460 18 31 level level NOUN fcis-1460 18 32 feature feature NOUN fcis-1460 18 33 information information NOUN fcis-1460 18 34 . . PUNCT fcis-1460 19 1 at at ADP fcis-1460 19 2 the the DET fcis-1460 19 3 same same ADJ fcis-1460 19 4 time time NOUN fcis-1460 19 5 , , PUNCT fcis-1460 19 6 the the DET fcis-1460 19 7 problem problem NOUN fcis-1460 19 8 of of ADP fcis-1460 19 9 structural structural ADJ fcis-1460 19 10 redundancy redundancy NOUN fcis-1460 19 11 is be AUX fcis-1460 19 12 also also ADV fcis-1460 19 13 solved solve VERB fcis-1460 19 14 . . PUNCT fcis-1460 20 1 although although SCONJ fcis-1460 20 2 the the DET fcis-1460 20 3 cnn cnn PROPN fcis-1460 20 4 - - PUNCT fcis-1460 20 5 based base VERB fcis-1460 20 6 algorithm algorithm NOUN fcis-1460 20 7 has have VERB fcis-1460 20 8 a a DET fcis-1460 20 9 high high ADJ fcis-1460 20 10 accuracy accuracy NOUN fcis-1460 20 11 , , PUNCT fcis-1460 20 12 it it PRON fcis-1460 20 13 is be AUX fcis-1460 20 14 common common ADJ fcis-1460 20 15 that that SCONJ fcis-1460 20 16 the the DET fcis-1460 20 17 extracted extract VERB fcis-1460 20 18 features feature NOUN fcis-1460 20 19 lose lose VERB fcis-1460 20 20 a a DET fcis-1460 20 21 lot lot NOUN fcis-1460 20 22 of of ADP fcis-1460 20 23 spatial spatial ADJ fcis-1460 20 24 information information NOUN fcis-1460 20 25 due due ADP fcis-1460 20 26 to to ADP fcis-1460 20 27 the the DET fcis-1460 20 28 pooling pool VERB fcis-1460 20 29 layer layer NOUN fcis-1460 20 30 . . PUNCT fcis-1460 21 1 eventually eventually ADV fcis-1460 21 2 , , PUNCT fcis-1460 21 3 the the DET fcis-1460 21 4 semantic semantic ADJ fcis-1460 21 5 segmentation segmentation NOUN fcis-1460 21 6 network network NOUN fcis-1460 21 7 structure structure NOUN fcis-1460 21 8 redundancy redundancy NOUN fcis-1460 21 9 , , PUNCT fcis-1460 21 10 large large ADJ fcis-1460 21 11 amount amount NOUN fcis-1460 21 12 of of ADP fcis-1460 21 13 computation computation NOUN fcis-1460 21 14 , , PUNCT fcis-1460 21 15 segmentation segmentation NOUN fcis-1460 21 16 errors error NOUN fcis-1460 21 17 and and CCONJ fcis-1460 21 18 other other ADJ fcis-1460 21 19 problems problem NOUN fcis-1460 21 20 appear appear VERB fcis-1460 21 21 . . PUNCT fcis-1460 22 1 here here ADV fcis-1460 22 2 we we PRON fcis-1460 22 3 discrete discrete VERB fcis-1460 22 4 the the DET fcis-1460 22 5 statistical statistical ADJ fcis-1460 22 6 properties property NOUN fcis-1460 22 7 and and CCONJ fcis-1460 22 8 distribution distribution NOUN fcis-1460 22 9 properties property NOUN fcis-1460 22 10 of of ADP fcis-1460 22 11 the the DET fcis-1460 22 12 data datum NOUN fcis-1460 22 13 to to ADP fcis-1460 22 14 the the DET fcis-1460 22 15 data data NOUN fcis-1460 22 16 distribution distribution NOUN fcis-1460 22 17 of of ADP fcis-1460 22 18 image image NOUN fcis-1460 22 19 edge edge NOUN fcis-1460 22 20 features feature NOUN fcis-1460 22 21 , , PUNCT fcis-1460 22 22 and and CCONJ fcis-1460 22 23 design design VERB fcis-1460 22 24 the the DET fcis-1460 22 25 discrete discrete ADJ fcis-1460 22 26 network network NOUN fcis-1460 22 27 discnet discnet NOUN fcis-1460 22 28 . . PUNCT fcis-1460 23 1 discnet discnet NOUN fcis-1460 23 2 extracts extract VERB fcis-1460 23 3 the the DET fcis-1460 23 4 features feature NOUN fcis-1460 23 5 with with ADP fcis-1460 23 6 larger large ADJ fcis-1460 23 7 discreteness discreteness NOUN fcis-1460 23 8 in in ADP fcis-1460 23 9 the the DET fcis-1460 23 10 image image NOUN fcis-1460 23 11 by by ADP fcis-1460 23 12 analyzing analyze VERB fcis-1460 23 13 the the DET fcis-1460 23 14 distribution distribution NOUN fcis-1460 23 15 characteristics characteristic NOUN fcis-1460 23 16 of of ADP fcis-1460 23 17 discrete discrete ADJ fcis-1460 23 18 data datum NOUN fcis-1460 23 19 , , PUNCT fcis-1460 23 20 that that ADV fcis-1460 23 21 is is ADV fcis-1460 23 22 , , PUNCT fcis-1460 23 23 edge edge NOUN fcis-1460 23 24 features feature NOUN fcis-1460 23 25 . . PUNCT fcis-1460 24 1 after after SCONJ fcis-1460 24 2 discnet discnet NOUN fcis-1460 24 3 extracts extract VERB fcis-1460 24 4 the the DET fcis-1460 24 5 edge edge NOUN fcis-1460 24 6 features feature NOUN fcis-1460 24 7 of of ADP fcis-1460 24 8 the the DET fcis-1460 24 9 image image NOUN fcis-1460 24 10 , , PUNCT fcis-1460 24 11 the the DET fcis-1460 24 12 edge edge NOUN fcis-1460 24 13 features feature NOUN fcis-1460 24 14 are be AUX fcis-1460 24 15 regressed regress VERB fcis-1460 24 16 to to PART fcis-1460 24 17 locate locate VERB fcis-1460 24 18 the the DET fcis-1460 24 19 edge edge NOUN fcis-1460 24 20 contour contour NOUN fcis-1460 24 21 of of ADP fcis-1460 24 22 the the DET fcis-1460 24 23 target target NOUN fcis-1460 24 24 , , PUNCT fcis-1460 24 25 and and CCONJ fcis-1460 24 26 then then ADV fcis-1460 24 27 achieve achieve VERB fcis-1460 24 28 accurate accurate ADJ fcis-1460 24 29 semantic semantic ADJ fcis-1460 24 30 segmentation segmentation NOUN fcis-1460 24 31 . . PUNCT fcis-1460 25 1 under under ADP fcis-1460 25 2 the the DET fcis-1460 25 3 cityscapes cityscape NOUN fcis-1460 25 4 data datum NOUN fcis-1460 25 5 set set NOUN fcis-1460 25 6 , , PUNCT fcis-1460 25 7 discnet discnet NOUN fcis-1460 25 8 and and CCONJ fcis-1460 25 9 sota sota PROPN fcis-1460 25 10 algorithm algorithm NOUN fcis-1460 25 11 are be AUX fcis-1460 25 12 compared compare VERB fcis-1460 25 13 . . PUNCT fcis-1460 26 1 the the DET fcis-1460 26 2 experimental experimental ADJ fcis-1460 26 3 results result NOUN fcis-1460 26 4 show show VERB fcis-1460 26 5 that that SCONJ fcis-1460 26 6 discnet discnet NOUN fcis-1460 26 7 has have VERB fcis-1460 26 8 certain certain ADJ fcis-1460 26 9 advantages advantage NOUN fcis-1460 26 10 in in ADP fcis-1460 26 11 terms term NOUN fcis-1460 26 12 of of ADP fcis-1460 26 13 accuracy accuracy NOUN fcis-1460 26 14 and and CCONJ fcis-1460 26 15 speed speed NOUN fcis-1460 26 16 . . PUNCT fcis-1460 27 1 2 2 X fcis-1460 27 2 . . X fcis-1460 27 3 method method VERB fcis-1460 27 4 the the DET fcis-1460 27 5 input input NOUN fcis-1460 27 6 feature feature NOUN fcis-1460 27 7 passes pass VERB fcis-1460 27 8 through through ADP fcis-1460 27 9 the the DET fcis-1460 27 10 sliding slide VERB fcis-1460 27 11 window window NOUN fcis-1460 27 12 to to PART fcis-1460 27 13 obtain obtain VERB fcis-1460 27 14 n n PRON fcis-1460 27 15 feature feature NOUN fcis-1460 27 16 maps map NOUN fcis-1460 27 17 pi pi NOUN fcis-1460 27 18 ( ( PUNCT fcis-1460 27 19 i=1 i=1 PROPN fcis-1460 27 20 , , PUNCT fcis-1460 27 21 2 2 NUM fcis-1460 27 22 ... ... PUNCT fcis-1460 27 23 , , PUNCT fcis-1460 27 24 n n CCONJ fcis-1460 27 25 ) ) PUNCT fcis-1460 27 26 with with ADP fcis-1460 27 27 the the DET fcis-1460 27 28 same same ADJ fcis-1460 27 29 dimension dimension NOUN fcis-1460 27 30 h*w h*w ADJ fcis-1460 27 31 and and CCONJ fcis-1460 27 32 different different ADJ fcis-1460 27 33 eigenvalues eigenvalue NOUN fcis-1460 27 34 . . PUNCT fcis-1460 28 1 through through ADP fcis-1460 28 2 formula formula NOUN fcis-1460 28 3 1 1 NUM fcis-1460 28 4 , , PUNCT fcis-1460 28 5 we we PRON fcis-1460 28 6 can can AUX fcis-1460 28 7 obtain obtain VERB fcis-1460 28 8 the the DET fcis-1460 28 9 correlation correlation NOUN fcis-1460 28 10 coefficient coefficient NOUN fcis-1460 28 11 between between ADP fcis-1460 28 12 the the DET fcis-1460 28 13 feature feature NOUN fcis-1460 28 14 maps map NOUN fcis-1460 28 15 pm pm NOUN fcis-1460 28 16 and and CCONJ fcis-1460 28 17 pm+1 pm+1 VERB fcis-1460 28 18 , , PUNCT fcis-1460 28 19 expressing express VERB fcis-1460 28 20 the the DET fcis-1460 28 21 correlation correlation NOUN fcis-1460 28 22 between between ADP fcis-1460 28 23 the the DET fcis-1460 28 24 two two NUM fcis-1460 28 25 feature feature NOUN fcis-1460 28 26 maps map NOUN fcis-1460 28 27 . . PUNCT fcis-1460 29 1 the the DET fcis-1460 29 2 eigenvalues eigenvalues PROPN fcis-1460 29 3 ri ri PROPN fcis-1460 29 4 , , PUNCT fcis-1460 29 5 j j PROPN fcis-1460 29 6 of of ADP fcis-1460 29 7 the the DET fcis-1460 29 8 last last ADJ fcis-1460 29 9 n n NOUN fcis-1460 29 10 positions position NOUN fcis-1460 29 11 forms form VERB fcis-1460 29 12 a a DET fcis-1460 29 13 new new ADJ fcis-1460 29 14 feature feature NOUN fcis-1460 29 15 map map NOUN fcis-1460 29 16 r r NOUN fcis-1460 29 17 , , PUNCT fcis-1460 29 18 which which PRON fcis-1460 29 19 ensures ensure VERB fcis-1460 29 20 the the DET fcis-1460 29 21 correlation correlation NOUN fcis-1460 29 22 between between ADP fcis-1460 29 23 the the DET fcis-1460 29 24 eigenvalues eigenvalue NOUN fcis-1460 29 25 . . PUNCT fcis-1460 29 26 π‘Ÿ(𝑖,𝑗 π‘Ÿ(𝑖,𝑗 NOUN fcis-1460 29 27 ) ) PUNCT fcis-1460 30 1 = = SYM fcis-1460 30 2 βˆ‘(π‘ƒπ‘š|(𝑖 βˆ‘(π‘ƒπ‘š|(𝑖 NOUN fcis-1460 30 3 , , PUNCT fcis-1460 30 4 𝑗 𝑗 NOUN fcis-1460 30 5 ) ) PUNCT fcis-1460 30 6 βˆ’ βˆ’ NOUN fcis-1460 30 7 π‘ƒπ‘š|(𝑖 π‘ƒπ‘š|(𝑖 NOUN fcis-1460 30 8 , , PUNCT fcis-1460 30 9 𝑗))(𝑃 𝑗))(𝑃 PROPN fcis-1460 30 10 βˆ’ βˆ’ NOUN fcis-1460 30 11 π‘ƒπ‘š+1 π‘ƒπ‘š+1 NOUN fcis-1460 30 12 ) ) PUNCT fcis-1460 30 13 √(βˆ‘(π‘ƒπ‘š|(𝑖 √(βˆ‘(π‘ƒπ‘š|(𝑖 NOUN fcis-1460 30 14 , , PUNCT fcis-1460 30 15 𝑗 𝑗 NOUN fcis-1460 30 16 ) ) PUNCT fcis-1460 30 17 βˆ’ βˆ’ NOUN fcis-1460 30 18 π‘ƒπ‘š|(𝑖 π‘ƒπ‘š|(𝑖 NOUN fcis-1460 30 19 , , PUNCT fcis-1460 30 20 𝑗))2)(βˆ‘(π‘ƒπ‘š+1 𝑗))2)(βˆ‘(π‘ƒπ‘š+1 PROPN fcis-1460 30 21 βˆ’ βˆ’ NUM fcis-1460 30 22 π‘ƒπ‘š+1)2 π‘ƒπ‘š+1)2 ADJ fcis-1460 30 23 ) ) PUNCT fcis-1460 30 24 ( ( PUNCT fcis-1460 30 25 1 1 X fcis-1460 30 26 ) ) PUNCT fcis-1460 30 27 n n NOUN fcis-1460 30 28 = = SYM fcis-1460 30 29 ⌈ ⌈ PROPN fcis-1460 30 30 𝐻 𝐻 PRON fcis-1460 30 31 βˆ’ βˆ’ NOUN fcis-1460 30 32 β„Ž β„Ž NOUN fcis-1460 30 33 + + CCONJ fcis-1460 30 34 2 2 NUM fcis-1460 30 35 βˆ— βˆ— NOUN fcis-1460 30 36 𝑝 𝑝 NOUN fcis-1460 30 37 𝑠 𝑠 PROPN fcis-1460 30 38 + + NUM fcis-1460 30 39 1βŒ‰ 1βŒ‰ PROPN fcis-1460 30 40 βˆ— βˆ— NOUN fcis-1460 30 41 ⌈ ⌈ NOUN fcis-1460 30 42 π‘Š π‘Š PROPN fcis-1460 30 43 βˆ’ βˆ’ NOUN fcis-1460 30 44 𝑀 𝑀 ADP fcis-1460 30 45 + + CCONJ fcis-1460 30 46 2 2 NUM fcis-1460 30 47 βˆ— βˆ— NOUN fcis-1460 30 48 𝑝 𝑝 NOUN fcis-1460 30 49 𝑠 𝑠 PROPN fcis-1460 30 50 + + NUM fcis-1460 30 51 1βŒ‰ 1βŒ‰ PROPN fcis-1460 30 52 ( ( PUNCT fcis-1460 30 53 2 2 NUM fcis-1460 30 54 ) ) PUNCT fcis-1460 30 55 r r NOUN fcis-1460 30 56 = = PUNCT fcis-1460 30 57 π‘Ÿ π‘Ÿ NOUN fcis-1460 30 58 βˆ— βˆ— NOUN fcis-1460 30 59 𝑃 𝑃 NOUN fcis-1460 30 60 ( ( PUNCT fcis-1460 30 61 3 3 NUM fcis-1460 30 62 ) ) PUNCT fcis-1460 30 63 pm|(i pm|(i ADJ fcis-1460 30 64 , , PUNCT fcis-1460 30 65 j j NOUN fcis-1460 30 66 ) ) PUNCT fcis-1460 30 67 is be AUX fcis-1460 30 68 the the DET fcis-1460 30 69 feature feature NOUN fcis-1460 30 70 map map NOUN fcis-1460 30 71 with with ADP fcis-1460 30 72 point point NOUN fcis-1460 30 73 ( ( PUNCT fcis-1460 30 74 i i PROPN fcis-1460 30 75 , , PUNCT fcis-1460 30 76 j j PROPN fcis-1460 30 77 ) ) PUNCT fcis-1460 30 78 as as ADP fcis-1460 30 79 the the DET fcis-1460 30 80 upper upper ADJ fcis-1460 30 81 left left NOUN fcis-1460 30 82 corner corner NOUN fcis-1460 30 83 , , PUNCT fcis-1460 30 84 π‘ƒπ‘š|(𝑖 π‘ƒπ‘š|(𝑖 NOUN fcis-1460 30 85 , , PUNCT fcis-1460 30 86 𝑗)is 𝑗)is PROPN fcis-1460 30 87 the the DET fcis-1460 30 88 mean mean NOUN fcis-1460 30 89 of of ADP fcis-1460 30 90 the the DET fcis-1460 30 91 feature feature NOUN fcis-1460 30 92 map map NOUN fcis-1460 30 93 pm pm NOUN fcis-1460 30 94 . . PUNCT fcis-1460 31 1 pm+1 pm+1 PROPN fcis-1460 31 2 is be AUX fcis-1460 31 3 the the DET fcis-1460 31 4 next next ADJ fcis-1460 31 5 feature feature NOUN fcis-1460 31 6 map map NOUN fcis-1460 31 7 of of ADP fcis-1460 31 8 the the DET fcis-1460 31 9 pm pm NOUN fcis-1460 31 10 . . PUNCT fcis-1460 32 1 then then ADV fcis-1460 32 2 , , PUNCT fcis-1460 32 3 the the DET fcis-1460 32 4 feature feature NOUN fcis-1460 32 5 map map NOUN fcis-1460 32 6 r r NOUN fcis-1460 32 7 is be AUX fcis-1460 32 8 dotmultiplied dotmultiplie VERB fcis-1460 32 9 with with ADP fcis-1460 32 10 the the DET fcis-1460 32 11 p p NOUN fcis-1460 32 12 , , PUNCT fcis-1460 32 13 and and CCONJ fcis-1460 32 14 finally finally ADV fcis-1460 32 15 the the DET fcis-1460 32 16 feature feature NOUN fcis-1460 32 17 map map NOUN fcis-1460 32 18 r r NOUN fcis-1460 32 19 pooled pool VERB fcis-1460 32 20 by by ADP fcis-1460 32 21 the the DET fcis-1460 32 22 rel rel NOUN fcis-1460 32 23 layer layer NOUN fcis-1460 32 24 is be AUX fcis-1460 32 25 obtained obtain VERB fcis-1460 32 26 . . PUNCT fcis-1460 33 1 table table NOUN fcis-1460 33 2 1 1 NUM fcis-1460 33 3 . . PUNCT fcis-1460 34 1 the the DET fcis-1460 34 2 network network NOUN fcis-1460 34 3 structure structure NOUN fcis-1460 34 4 parameters parameter NOUN fcis-1460 34 5 are be AUX fcis-1460 34 6 shown show VERB fcis-1460 34 7 in in ADP fcis-1460 34 8 the the DET fcis-1460 34 9 following follow VERB fcis-1460 34 10 table table NOUN fcis-1460 34 11 layer layer NOUN fcis-1460 34 12 kernel kernel NOUN fcis-1460 34 13 / / SYM fcis-1460 34 14 stride stride NOUN fcis-1460 34 15 parameters parameter NOUN fcis-1460 34 16 conv1 conv1 VERB fcis-1460 34 17 128 128 NUM fcis-1460 34 18 * * SYM fcis-1460 34 19 11 11 NUM fcis-1460 34 20 * * SYM fcis-1460 34 21 11/4 11/4 NUM fcis-1460 34 22 15488 15488 NUM fcis-1460 34 23 dis dis NOUN fcis-1460 34 24 pool pool VERB fcis-1460 34 25 3 3 NUM fcis-1460 34 26 * * SYM fcis-1460 34 27 3/1 3/1 NUM fcis-1460 34 28 conv3/5/7/9/11/13 conv3/5/7/9/11/13 ADJ fcis-1460 34 29 8 8 NUM fcis-1460 34 30 * * SYM fcis-1460 34 31 1 1 NUM fcis-1460 34 32 * * SYM fcis-1460 34 33 3/2 3/2 NUM fcis-1460 34 34 3072 3072 NUM fcis-1460 34 35 8 8 NUM fcis-1460 34 36 * * SYM fcis-1460 34 37 3 3 NUM fcis-1460 34 38 * * NUM fcis-1460 34 39 1/2 1/2 NUM fcis-1460 34 40 rel rel NOUN fcis-1460 34 41 pool pool NOUN fcis-1460 34 42 3 3 NUM fcis-1460 34 43 * * SYM fcis-1460 34 44 3/2 3/2 NUM fcis-1460 34 45 conv4/8/12 conv4/8/12 NOUN fcis-1460 34 46 16 16 NUM fcis-1460 34 47 * * SYM fcis-1460 34 48 3 3 NUM fcis-1460 34 49 * * SYM fcis-1460 34 50 3/2 3/2 NUM fcis-1460 34 51 1152 1152 NUM fcis-1460 34 52 conv2/6/10 conv2/6/10 ADJ fcis-1460 34 53 64 64 NUM fcis-1460 34 54 * * NUM fcis-1460 34 55 11 11 NUM fcis-1460 34 56 * * SYM fcis-1460 34 57 11/4 11/4 NUM fcis-1460 34 58 61952 61952 NUM fcis-1460 35 1 the the DET fcis-1460 35 2 input input NOUN fcis-1460 35 3 data data NOUN fcis-1460 35 4 is be AUX fcis-1460 35 5 multiplied multiply VERB fcis-1460 35 6 by by ADP fcis-1460 35 7 the the DET fcis-1460 35 8 discrete discrete ADJ fcis-1460 35 9 coefficient coefficient NOUN fcis-1460 35 10 obtained obtain VERB fcis-1460 35 11 by by ADP fcis-1460 35 12 the the DET fcis-1460 35 13 dis dis PROPN fcis-1460 35 14 pooling pool VERB fcis-1460 35 15 layer layer NOUN fcis-1460 35 16 . . PUNCT fcis-1460 36 1 the the DET fcis-1460 36 2 purpose purpose NOUN fcis-1460 36 3 is be AUX fcis-1460 36 4 to to PART fcis-1460 36 5 calibrate calibrate VERB fcis-1460 36 6 the the DET fcis-1460 36 7 spatial spatial ADJ fcis-1460 36 8 position position NOUN fcis-1460 36 9 of of ADP fcis-1460 36 10 the the DET fcis-1460 36 11 edge edge NOUN fcis-1460 36 12 feature feature NOUN fcis-1460 36 13 through through ADP fcis-1460 36 14 the the DET fcis-1460 36 15 discrete discrete ADJ fcis-1460 36 16 coefficient coefficient NOUN fcis-1460 36 17 . . PUNCT fcis-1460 37 1 the the DET fcis-1460 37 2 result result NOUN fcis-1460 37 3 obtained obtain VERB fcis-1460 37 4 here here ADV fcis-1460 37 5 is be AUX fcis-1460 37 6 then then ADV fcis-1460 37 7 added add VERB fcis-1460 37 8 to to ADP fcis-1460 37 9 the the DET fcis-1460 37 10 input input NOUN fcis-1460 37 11 data datum NOUN fcis-1460 37 12 . . PUNCT fcis-1460 38 1 the the DET fcis-1460 38 2 purpose purpose NOUN fcis-1460 38 3 it it PRON fcis-1460 38 4 enhances enhance VERB fcis-1460 38 5 the the DET fcis-1460 38 6 information information NOUN fcis-1460 38 7 of of ADP fcis-1460 38 8 edge edge NOUN fcis-1460 38 9 features feature NOUN fcis-1460 38 10 on on ADP fcis-1460 38 11 the the DET fcis-1460 38 12 one one NUM fcis-1460 38 13 hand hand NOUN fcis-1460 38 14 , , PUNCT fcis-1460 38 15 and and CCONJ fcis-1460 38 16 preserves preserve VERB fcis-1460 38 17 the the DET fcis-1460 38 18 correlation correlation NOUN fcis-1460 38 19 between between ADP fcis-1460 38 20 features feature NOUN fcis-1460 38 21 on on ADP fcis-1460 38 22 the the DET fcis-1460 38 23 other other ADJ fcis-1460 38 24 hand hand NOUN fcis-1460 38 25 . . PUNCT fcis-1460 39 1 therefore therefore ADV fcis-1460 39 2 , , PUNCT fcis-1460 39 3 the the DET fcis-1460 39 4 point point NOUN fcis-1460 39 5 multiplication multiplication NOUN fcis-1460 39 6 is be AUX fcis-1460 39 7 to to PART fcis-1460 39 8 calibrate calibrate VERB fcis-1460 39 9 the the DET fcis-1460 39 10 spatial spatial ADJ fcis-1460 39 11 position position NOUN fcis-1460 39 12 of of ADP fcis-1460 39 13 the the DET fcis-1460 39 14 edge edge NOUN fcis-1460 39 15 features feature NOUN fcis-1460 39 16 , , PUNCT fcis-1460 39 17 and and CCONJ fcis-1460 39 18 the the DET fcis-1460 39 19 46 46 NUM fcis-1460 39 20 purpose purpose NOUN fcis-1460 39 21 of of ADP fcis-1460 39 22 addition addition NOUN fcis-1460 39 23 is be AUX fcis-1460 39 24 to to PART fcis-1460 39 25 strengthen strengthen VERB fcis-1460 39 26 the the DET fcis-1460 39 27 information information NOUN fcis-1460 39 28 of of ADP fcis-1460 39 29 the the DET fcis-1460 39 30 edge edge NOUN fcis-1460 39 31 features feature NOUN fcis-1460 39 32 and and CCONJ fcis-1460 39 33 preserve preserve VERB fcis-1460 39 34 the the DET fcis-1460 39 35 correlation correlation NOUN fcis-1460 39 36 between between ADP fcis-1460 39 37 the the DET fcis-1460 39 38 features feature NOUN fcis-1460 39 39 . . PUNCT fcis-1460 40 1 the the DET fcis-1460 40 2 discnet discnet NOUN fcis-1460 40 3 structure structure NOUN fcis-1460 40 4 refers refer VERB fcis-1460 40 5 to to ADP fcis-1460 40 6 the the DET fcis-1460 40 7 residual residual ADJ fcis-1460 40 8 effect effect NOUN fcis-1460 40 9 of of ADP fcis-1460 40 10 resnet resnet NOUN fcis-1460 40 11 , , PUNCT fcis-1460 40 12 which which PRON fcis-1460 40 13 effectively effectively ADV fcis-1460 40 14 extracts extract VERB fcis-1460 40 15 edge edge NOUN fcis-1460 40 16 features feature NOUN fcis-1460 40 17 on on ADP fcis-1460 40 18 the the DET fcis-1460 40 19 one one NUM fcis-1460 40 20 hand hand NOUN fcis-1460 40 21 , , PUNCT fcis-1460 40 22 and and CCONJ fcis-1460 40 23 makes make VERB fcis-1460 40 24 the the DET fcis-1460 40 25 model model NOUN fcis-1460 40 26 more more ADV fcis-1460 40 27 lightweight lightweight NOUN fcis-1460 40 28 on on ADP fcis-1460 40 29 the the DET fcis-1460 40 30 other other ADJ fcis-1460 40 31 hand hand NOUN fcis-1460 40 32 . . PUNCT fcis-1460 41 1 3 3 X fcis-1460 41 2 . . X fcis-1460 41 3 experiments experiment NOUN fcis-1460 41 4 3.1 3.1 NUM fcis-1460 41 5 . . PUNCT fcis-1460 42 1 dataset dataset NOUN fcis-1460 42 2 cityscapes cityscape NOUN fcis-1460 42 3 contains contain VERB fcis-1460 42 4 a a DET fcis-1460 42 5 total total NOUN fcis-1460 42 6 of of ADP fcis-1460 42 7 5000 5000 NUM fcis-1460 42 8 fine fine ADJ fcis-1460 42 9 images image NOUN fcis-1460 42 10 , , PUNCT fcis-1460 42 11 of of ADP fcis-1460 42 12 which which PRON fcis-1460 42 13 2975 2975 NUM fcis-1460 42 14 are be AUX fcis-1460 42 15 training training NOUN fcis-1460 42 16 images image NOUN fcis-1460 42 17 , , PUNCT fcis-1460 42 18 500 500 NUM fcis-1460 42 19 validation validation NOUN fcis-1460 42 20 images image NOUN fcis-1460 42 21 and and CCONJ fcis-1460 42 22 1525 1525 NUM fcis-1460 42 23 testing testing NOUN fcis-1460 42 24 images image NOUN fcis-1460 42 25 . . PUNCT fcis-1460 43 1 in in ADP fcis-1460 43 2 addition addition NOUN fcis-1460 43 3 , , PUNCT fcis-1460 43 4 the the DET fcis-1460 43 5 dataset dataset NOUN fcis-1460 43 6 contains contain VERB fcis-1460 43 7 20k 20k NOUN fcis-1460 43 8 roughly roughly ADV fcis-1460 43 9 annotated annotate VERB fcis-1460 43 10 images image NOUN fcis-1460 43 11 . . PUNCT fcis-1460 44 1 3.2 3.2 NUM fcis-1460 44 2 . . PUNCT fcis-1460 44 3 comparison comparison NOUN fcis-1460 44 4 with with ADP fcis-1460 44 5 sota sota NOUN fcis-1460 44 6 we we PRON fcis-1460 44 7 first first ADV fcis-1460 44 8 trim trim VERB fcis-1460 44 9 the the DET fcis-1460 44 10 images image NOUN fcis-1460 44 11 of of ADP fcis-1460 44 12 cistyscapes cistyscape NOUN fcis-1460 44 13 datasets dataset NOUN fcis-1460 44 14 to to ADP fcis-1460 44 15 a a DET fcis-1460 44 16 size size NOUN fcis-1460 44 17 of of ADP fcis-1460 44 18 500 500 NUM fcis-1460 44 19 * * SYM fcis-1460 44 20 500 500 NUM fcis-1460 44 21 , , PUNCT fcis-1460 44 22 and and CCONJ fcis-1460 44 23 set set VERB fcis-1460 44 24 the the DET fcis-1460 44 25 initial initial ADJ fcis-1460 44 26 learning learning NOUN fcis-1460 44 27 rate rate NOUN fcis-1460 44 28 to to ADP fcis-1460 44 29 1*e-5 1*e-5 NUM fcis-1460 44 30 and and CCONJ fcis-1460 44 31 the the DET fcis-1460 44 32 epoch epoch NOUN fcis-1460 44 33 to to ADP fcis-1460 44 34 12000 12000 NUM fcis-1460 44 35 . . PUNCT fcis-1460 45 1 the the DET fcis-1460 45 2 cityscapes cityscape NOUN fcis-1460 45 3 dataset dataset NOUN fcis-1460 45 4 has have VERB fcis-1460 45 5 2975 2975 NUM fcis-1460 45 6 training training NOUN fcis-1460 45 7 images image NOUN fcis-1460 45 8 and and CCONJ fcis-1460 45 9 1525 1525 NUM fcis-1460 45 10 testing testing NOUN fcis-1460 45 11 images image NOUN fcis-1460 45 12 . . PUNCT fcis-1460 46 1 the the DET fcis-1460 46 2 training training NOUN fcis-1460 46 3 platforms platform NOUN fcis-1460 46 4 are be AUX fcis-1460 46 5 ryzen ryzen ADJ fcis-1460 46 6 7 7 NUM fcis-1460 46 7 3800x 3800x NUM fcis-1460 46 8 and and CCONJ fcis-1460 46 9 rtx rtx PROPN fcis-1460 46 10 2070 2070 NUM fcis-1460 46 11 . . PUNCT fcis-1460 47 1 the the DET fcis-1460 47 2 optimization optimization NOUN fcis-1460 47 3 function function NOUN fcis-1460 47 4 is be AUX fcis-1460 47 5 adam adam PROPN fcis-1460 47 6 optimizer optimizer NOUN fcis-1460 47 7 . . PUNCT fcis-1460 48 1 the the DET fcis-1460 48 2 loss loss NOUN fcis-1460 48 3 function function NOUN fcis-1460 48 4 uses use VERB fcis-1460 48 5 formula formula NOUN fcis-1460 48 6 4 4 NUM fcis-1460 48 7 to to PART fcis-1460 48 8 calculate calculate VERB fcis-1460 48 9 the the DET fcis-1460 48 10 error error NOUN fcis-1460 48 11 between between ADP fcis-1460 48 12 the the DET fcis-1460 48 13 true true ADJ fcis-1460 48 14 value value NOUN fcis-1460 48 15 and and CCONJ fcis-1460 48 16 the the DET fcis-1460 48 17 predicted predict VERB fcis-1460 48 18 value value NOUN fcis-1460 48 19 , , PUNCT fcis-1460 48 20 and and CCONJ fcis-1460 48 21 uses use VERB fcis-1460 48 22 iou iou VERB fcis-1460 48 23 to to PART fcis-1460 48 24 evaluate evaluate VERB fcis-1460 48 25 the the DET fcis-1460 48 26 test test NOUN fcis-1460 48 27 results result NOUN fcis-1460 48 28 . . PUNCT fcis-1460 49 1 yp yp PROPN fcis-1460 49 2 and and CCONJ fcis-1460 49 3 yt yt PROPN fcis-1460 49 4 represent represent VERB fcis-1460 49 5 the the DET fcis-1460 49 6 predicted predict VERB fcis-1460 49 7 values value NOUN fcis-1460 49 8 and and CCONJ fcis-1460 49 9 actual actual ADJ fcis-1460 49 10 values value NOUN fcis-1460 49 11 respectively respectively ADV fcis-1460 49 12 . . PUNCT fcis-1460 50 1 dice(p dice(p NOUN fcis-1460 50 2 , , PUNCT fcis-1460 50 3 t t PROPN fcis-1460 50 4 ) ) PUNCT fcis-1460 50 5 = = SYM fcis-1460 50 6 2 2 NUM fcis-1460 50 7 βˆ— βˆ— NOUN fcis-1460 50 8 |𝑦𝑝 |𝑦𝑝 X fcis-1460 50 9 ∩ ∩ ADJ fcis-1460 50 10 𝑦𝑑| 𝑦𝑑| PROPN fcis-1460 50 11 ( ( PUNCT fcis-1460 50 12 |𝑦𝑝| |𝑦𝑝| ADJ fcis-1460 50 13 + + CCONJ fcis-1460 50 14 |𝑦𝑑|)⁄ |𝑦𝑑|)⁄ ADJ fcis-1460 50 15 ( ( PUNCT fcis-1460 50 16 4 4 X fcis-1460 50 17 ) ) PUNCT fcis-1460 50 18 here here ADV fcis-1460 50 19 we we PRON fcis-1460 50 20 compare compare VERB fcis-1460 50 21 resnet resnet NOUN fcis-1460 50 22 with with ADP fcis-1460 50 23 some some DET fcis-1460 50 24 sota sota ADJ fcis-1460 50 25 algorithms algorithm NOUN fcis-1460 50 26 , , PUNCT fcis-1460 50 27 such such ADJ fcis-1460 50 28 as as ADP fcis-1460 50 29 deep deep ADJ fcis-1460 50 30 snake snake NOUN fcis-1460 51 1 [ [ X fcis-1460 51 2 8 8 NUM fcis-1460 51 3 ] ] PUNCT fcis-1460 51 4 , , PUNCT fcis-1460 51 5 unet[9 unet[9 PROPN fcis-1460 51 6 ] ] X fcis-1460 51 7 , , PUNCT fcis-1460 51 8 panet[10 panet[10 PROPN fcis-1460 51 9 ] ] PUNCT fcis-1460 51 10 , , PUNCT fcis-1460 51 11 fcis[11 fcis[11 PROPN fcis-1460 51 12 ] ] PUNCT fcis-1460 51 13 , , PUNCT fcis-1460 51 14 ese[12 ese[12 PROPN fcis-1460 51 15 - - PUNCT fcis-1460 51 16 13 13 NUM fcis-1460 51 17 ] ] PUNCT fcis-1460 51 18 , , PUNCT fcis-1460 51 19 etc etc X fcis-1460 51 20 . . X fcis-1460 52 1 the the DET fcis-1460 52 2 results result NOUN fcis-1460 52 3 for for ADP fcis-1460 52 4 the the DET fcis-1460 52 5 cityscapes cityscape NOUN fcis-1460 52 6 datasets dataset NOUN fcis-1460 52 7 are be AUX fcis-1460 52 8 shown show VERB fcis-1460 52 9 in in ADP fcis-1460 52 10 table table NOUN fcis-1460 52 11 2 2 NUM fcis-1460 52 12 below below ADV fcis-1460 52 13 . . PUNCT fcis-1460 53 1 table table NOUN fcis-1460 53 2 2 2 NUM fcis-1460 53 3 . . PUNCT fcis-1460 53 4 comparison comparison NOUN fcis-1460 53 5 results result NOUN fcis-1460 53 6 of of ADP fcis-1460 53 7 cityscapes cityscape NOUN fcis-1460 53 8 datasets dataset NOUN fcis-1460 53 9 . . PUNCT fcis-1460 54 1 n n ADV fcis-1460 54 2 et et NOUN fcis-1460 54 3 w w NOUN fcis-1460 54 4 o o X fcis-1460 55 1 rk rk NOUN fcis-1460 56 1 d d NOUN fcis-1460 56 2 is be AUX fcis-1460 56 3 c c NOUN fcis-1460 56 4 n n PRON fcis-1460 56 5 et et NOUN fcis-1460 57 1 d d NOUN fcis-1460 57 2 ee ee ADP fcis-1460 57 3 p p PROPN fcis-1460 57 4 u u PROPN fcis-1460 57 5 n n ADP fcis-1460 57 6 et et NOUN fcis-1460 57 7 p p NOUN fcis-1460 57 8 a a PRON fcis-1460 57 9 n n ADP fcis-1460 57 10 et et NOUN fcis-1460 57 11 f f NOUN fcis-1460 57 12 c c PROPN fcis-1460 57 13 is be AUX fcis-1460 57 14 e e NOUN fcis-1460 57 15 s s PROPN fcis-1460 57 16 e e X fcis-1460 57 17 s s PROPN fcis-1460 57 18 eg eg NOUN fcis-1460 57 19 n n ADP fcis-1460 57 20 et et NOUN fcis-1460 57 21 a a DET fcis-1460 57 22 u u X fcis-1460 57 23 c c X fcis-1460 57 24 ( ( PUNCT fcis-1460 57 25 % % NOUN fcis-1460 57 26 ) ) PUNCT fcis-1460 57 27 3 3 NUM fcis-1460 57 28 8 8 NUM fcis-1460 57 29 .6 .6 NUM fcis-1460 57 30 3 3 NUM fcis-1460 57 31 7 7 NUM fcis-1460 57 32 .4 .4 NUM fcis-1460 57 33 3 3 NUM fcis-1460 57 34 8 8 NUM fcis-1460 57 35 .4 .4 NUM fcis-1460 57 36 3 3 NUM fcis-1460 57 37 6 6 NUM fcis-1460 57 38 .5 .5 NUM fcis-1460 57 39 2 2 NUM fcis-1460 57 40 9 9 NUM fcis-1460 57 41 .4 .4 NUM fcis-1460 57 42 1 1 NUM fcis-1460 57 43 8 8 NUM fcis-1460 57 44 .6 .6 NUM fcis-1460 57 45 1 1 NUM fcis-1460 57 46 3 3 NUM fcis-1460 58 1 .2 .2 NUM fcis-1460 58 2 f f PROPN fcis-1460 58 3 p p X fcis-1460 58 4 s s PROPN fcis-1460 58 5 1 1 NUM fcis-1460 58 6 8 8 NUM fcis-1460 58 7 .6 .6 NUM fcis-1460 58 8 4 4 NUM fcis-1460 58 9 .6 .6 NUM fcis-1460 58 10 8 8 NUM fcis-1460 58 11 .6 .6 NUM fcis-1460 58 12 7 7 NUM fcis-1460 58 13 .5 .5 NUM fcis-1460 58 14 1 1 NUM fcis-1460 58 15 0 0 NUM fcis-1460 59 1 .2 .2 NUM fcis-1460 59 2 1 1 NUM fcis-1460 59 3 8 8 NUM fcis-1460 59 4 .7 .7 NUM fcis-1460 59 5 1 1 NUM fcis-1460 59 6 6 6 NUM fcis-1460 59 7 .5 .5 NUM fcis-1460 59 8 judging judge VERB fcis-1460 59 9 from from ADP fcis-1460 59 10 the the DET fcis-1460 59 11 accuracy accuracy NOUN fcis-1460 59 12 results result NOUN fcis-1460 59 13 of of ADP fcis-1460 59 14 the the DET fcis-1460 59 15 cityscapes cityscape NOUN fcis-1460 59 16 datasets dataset NOUN fcis-1460 59 17 , , PUNCT fcis-1460 59 18 resnet resnet NOUN fcis-1460 59 19 can can AUX fcis-1460 59 20 achieve achieve VERB fcis-1460 59 21 a a DET fcis-1460 59 22 good good ADJ fcis-1460 59 23 result result NOUN fcis-1460 59 24 . . PUNCT fcis-1460 60 1 unet unet NOUN fcis-1460 60 2 and and CCONJ fcis-1460 60 3 panet panet NOUN fcis-1460 60 4 can can AUX fcis-1460 60 5 enhance enhance VERB fcis-1460 60 6 the the DET fcis-1460 60 7 features feature NOUN fcis-1460 60 8 of of ADP fcis-1460 60 9 corresponding correspond VERB fcis-1460 60 10 locations location NOUN fcis-1460 60 11 by by ADP fcis-1460 60 12 concatenating concatenate VERB fcis-1460 60 13 data datum NOUN fcis-1460 60 14 dimensions dimension NOUN fcis-1460 60 15 . . PUNCT fcis-1460 61 1 through through ADP fcis-1460 61 2 feature feature NOUN fcis-1460 61 3 splicing splicing NOUN fcis-1460 61 4 , , PUNCT fcis-1460 61 5 the the DET fcis-1460 61 6 high high ADJ fcis-1460 61 7 - - PUNCT fcis-1460 61 8 resolution resolution NOUN fcis-1460 61 9 features feature NOUN fcis-1460 61 10 are be AUX fcis-1460 61 11 enhanced enhance VERB fcis-1460 61 12 by by ADP fcis-1460 61 13 the the DET fcis-1460 61 14 low low ADJ fcis-1460 61 15 - - PUNCT fcis-1460 61 16 resolution resolution NOUN fcis-1460 61 17 features feature NOUN fcis-1460 61 18 , , PUNCT fcis-1460 61 19 so so CCONJ fcis-1460 61 20 the the DET fcis-1460 61 21 edge edge NOUN fcis-1460 61 22 features feature NOUN fcis-1460 61 23 can can AUX fcis-1460 61 24 be be AUX fcis-1460 61 25 accurately accurately ADV fcis-1460 61 26 segmented segment VERB fcis-1460 61 27 by by ADP fcis-1460 61 28 locating locate VERB fcis-1460 61 29 the the DET fcis-1460 61 30 enhanced enhanced ADJ fcis-1460 61 31 features feature NOUN fcis-1460 61 32 . . PUNCT fcis-1460 62 1 however however ADV fcis-1460 62 2 , , PUNCT fcis-1460 62 3 due due ADP fcis-1460 62 4 to to ADP fcis-1460 62 5 the the DET fcis-1460 62 6 problem problem NOUN fcis-1460 62 7 of of ADP fcis-1460 62 8 feature feature NOUN fcis-1460 62 9 loss loss NOUN fcis-1460 62 10 in in ADP fcis-1460 62 11 the the DET fcis-1460 62 12 pooling pool VERB fcis-1460 62 13 layer layer NOUN fcis-1460 62 14 , , PUNCT fcis-1460 62 15 the the DET fcis-1460 62 16 segmentation segmentation NOUN fcis-1460 62 17 position position NOUN fcis-1460 62 18 is be AUX fcis-1460 62 19 inaccurate inaccurate ADJ fcis-1460 62 20 . . PUNCT fcis-1460 63 1 fcis fcis NOUN fcis-1460 63 2 and and CCONJ fcis-1460 63 3 ese ese NOUN fcis-1460 63 4 will will AUX fcis-1460 63 5 be be AUX fcis-1460 63 6 trained train VERB fcis-1460 63 7 on on ADP fcis-1460 63 8 the the DET fcis-1460 63 9 basis basis NOUN fcis-1460 63 10 of of ADP fcis-1460 63 11 fully fully ADV fcis-1460 63 12 convolutional convolutional ADJ fcis-1460 63 13 network network NOUN fcis-1460 63 14 by by ADP fcis-1460 63 15 means mean NOUN fcis-1460 63 16 of of ADP fcis-1460 63 17 encode encode ADJ fcis-1460 63 18 - - PUNCT fcis-1460 63 19 decode decode NOUN fcis-1460 63 20 . . PUNCT fcis-1460 64 1 fcis fcis PROPN fcis-1460 64 2 can can AUX fcis-1460 64 3 effectively effectively ADV fcis-1460 64 4 avoid avoid VERB fcis-1460 64 5 the the DET fcis-1460 64 6 problem problem NOUN fcis-1460 64 7 of of ADP fcis-1460 64 8 inaccurate inaccurate ADJ fcis-1460 64 9 information information NOUN fcis-1460 64 10 caused cause VERB fcis-1460 64 11 by by ADP fcis-1460 64 12 the the DET fcis-1460 64 13 loss loss NOUN fcis-1460 64 14 of of ADP fcis-1460 64 15 feature feature NOUN fcis-1460 64 16 information information NOUN fcis-1460 64 17 in in ADP fcis-1460 64 18 the the DET fcis-1460 64 19 pooling pool VERB fcis-1460 64 20 layer layer NOUN fcis-1460 64 21 , , PUNCT fcis-1460 64 22 but but CCONJ fcis-1460 64 23 it it PRON fcis-1460 64 24 will will AUX fcis-1460 64 25 also also ADV fcis-1460 64 26 reduce reduce VERB fcis-1460 64 27 the the DET fcis-1460 64 28 calculation calculation NOUN fcis-1460 64 29 speed speed NOUN fcis-1460 64 30 due due ADP fcis-1460 64 31 to to ADP fcis-1460 64 32 the the DET fcis-1460 64 33 excessive excessive ADJ fcis-1460 64 34 number number NOUN fcis-1460 64 35 of of ADP fcis-1460 64 36 convolutional convolutional ADJ fcis-1460 64 37 layers layer NOUN fcis-1460 64 38 . . PUNCT fcis-1460 65 1 here here ADV fcis-1460 65 2 we we PRON fcis-1460 65 3 extract extract VERB fcis-1460 65 4 relevant relevant ADJ fcis-1460 65 5 feature feature NOUN fcis-1460 65 6 about about ADP fcis-1460 65 7 the the DET fcis-1460 65 8 edge edge NOUN fcis-1460 65 9 features feature NOUN fcis-1460 65 10 of of ADP fcis-1460 65 11 the the DET fcis-1460 65 12 image image NOUN fcis-1460 65 13 through through ADP fcis-1460 65 14 the the DET fcis-1460 65 15 rel rel NOUN fcis-1460 65 16 layer layer NOUN fcis-1460 65 17 . . PUNCT fcis-1460 66 1 resnet resnet PROPN fcis-1460 66 2 uses use VERB fcis-1460 66 3 these these DET fcis-1460 66 4 features feature NOUN fcis-1460 66 5 to to PART fcis-1460 66 6 enhance enhance VERB fcis-1460 66 7 the the DET fcis-1460 66 8 edge edge NOUN fcis-1460 66 9 feature feature NOUN fcis-1460 66 10 and and CCONJ fcis-1460 66 11 achieve achieve VERB fcis-1460 66 12 feature feature NOUN fcis-1460 66 13 positioning positioning NOUN fcis-1460 66 14 , , PUNCT fcis-1460 66 15 and and CCONJ fcis-1460 66 16 then then ADV fcis-1460 66 17 achieve achieve VERB fcis-1460 66 18 the the DET fcis-1460 66 19 effect effect NOUN fcis-1460 66 20 of of ADP fcis-1460 66 21 image image NOUN fcis-1460 66 22 segmentation segmentation NOUN fcis-1460 66 23 , , PUNCT fcis-1460 66 24 as as SCONJ fcis-1460 66 25 shown show VERB fcis-1460 66 26 in in ADP fcis-1460 66 27 figure figure NOUN fcis-1460 66 28 5 5 NUM fcis-1460 66 29 . . PUNCT fcis-1460 66 30 from from ADP fcis-1460 66 31 the the DET fcis-1460 66 32 comparison comparison NOUN fcis-1460 66 33 results result VERB fcis-1460 66 34 in in ADP fcis-1460 66 35 figure figure NOUN fcis-1460 66 36 1 1 NUM fcis-1460 66 37 , , PUNCT fcis-1460 66 38 it it PRON fcis-1460 66 39 can can AUX fcis-1460 66 40 be be AUX fcis-1460 66 41 directly directly ADV fcis-1460 66 42 seen see VERB fcis-1460 66 43 that that SCONJ fcis-1460 66 44 resnet resnet NOUN fcis-1460 66 45 can can AUX fcis-1460 66 46 accurately accurately ADV fcis-1460 66 47 segment segment VERB fcis-1460 66 48 the the DET fcis-1460 66 49 target target NOUN fcis-1460 66 50 edge edge NOUN fcis-1460 66 51 . . PUNCT fcis-1460 67 1 unet unet NOUN fcis-1460 67 2 and and CCONJ fcis-1460 67 3 deep deep ADJ fcis-1460 67 4 snake snake NOUN fcis-1460 67 5 can can AUX fcis-1460 67 6 not not PART fcis-1460 67 7 accurately accurately ADV fcis-1460 67 8 locate locate VERB fcis-1460 67 9 the the DET fcis-1460 67 10 fine fine ADJ fcis-1460 67 11 boundary boundary ADJ fcis-1460 67 12 contour contour NOUN fcis-1460 67 13 , , PUNCT fcis-1460 67 14 and and CCONJ fcis-1460 67 15 also also ADV fcis-1460 67 16 have have VERB fcis-1460 67 17 the the DET fcis-1460 67 18 problem problem NOUN fcis-1460 67 19 of of ADP fcis-1460 67 20 inaccurate inaccurate ADJ fcis-1460 67 21 segmentation segmentation NOUN fcis-1460 67 22 for for ADP fcis-1460 67 23 small small ADJ fcis-1460 67 24 volume volume NOUN fcis-1460 67 25 targets target NOUN fcis-1460 67 26 . . PUNCT fcis-1460 68 1 it it PRON fcis-1460 68 2 can can AUX fcis-1460 68 3 be be AUX fcis-1460 68 4 seen see VERB fcis-1460 68 5 from from ADP fcis-1460 68 6 the the DET fcis-1460 68 7 comparison comparison NOUN fcis-1460 68 8 results result NOUN fcis-1460 68 9 of of ADP fcis-1460 68 10 segmentation segmentation NOUN fcis-1460 68 11 renderings rendering NOUN fcis-1460 68 12 and and CCONJ fcis-1460 68 13 accuracy accuracy NOUN fcis-1460 68 14 that that PRON fcis-1460 68 15 resnet resnet NOUN fcis-1460 68 16 has have VERB fcis-1460 68 17 certain certain ADJ fcis-1460 68 18 advantages advantage NOUN fcis-1460 68 19 . . PUNCT fcis-1460 69 1 relnet relnet ADJ fcis-1460 69 2 unetdeepnetimage unetdeepnetimage NOUN fcis-1460 69 3 figure figure NOUN fcis-1460 69 4 1 1 NUM fcis-1460 69 5 . . PUNCT fcis-1460 69 6 image image NOUN fcis-1460 69 7 represent represent VERB fcis-1460 69 8 the the DET fcis-1460 69 9 original original ADJ fcis-1460 69 10 image image NOUN fcis-1460 69 11 . . PUNCT fcis-1460 70 1 and and CCONJ fcis-1460 70 2 resnet resnet NOUN fcis-1460 70 3 , , PUNCT fcis-1460 70 4 deepnet deepnet NOUN fcis-1460 70 5 , , PUNCT fcis-1460 70 6 unet unet NOUN fcis-1460 70 7 correspond correspond VERB fcis-1460 70 8 to to ADP fcis-1460 70 9 the the DET fcis-1460 70 10 segmentation segmentation NOUN fcis-1460 70 11 effect effect NOUN fcis-1460 70 12 image image NOUN fcis-1460 70 13 of of ADP fcis-1460 70 14 these these DET fcis-1460 70 15 algorithms algorithm NOUN fcis-1460 70 16 respectively respectively ADV fcis-1460 70 17 . . PUNCT fcis-1460 71 1 4 4 X fcis-1460 71 2 . . X fcis-1460 71 3 conclusion conclusion NOUN fcis-1460 71 4 in in ADP fcis-1460 71 5 this this DET fcis-1460 71 6 paper paper NOUN fcis-1460 71 7 , , PUNCT fcis-1460 71 8 we we PRON fcis-1460 71 9 extract extract VERB fcis-1460 71 10 discrete discrete ADJ fcis-1460 71 11 features feature NOUN fcis-1460 71 12 in in ADP fcis-1460 71 13 the the DET fcis-1460 71 14 image image NOUN fcis-1460 71 15 by by ADP fcis-1460 71 16 analyzing analyze VERB fcis-1460 71 17 the the DET fcis-1460 71 18 statistical statistical ADJ fcis-1460 71 19 characteristics characteristic NOUN fcis-1460 71 20 of of ADP fcis-1460 71 21 discrete discrete ADJ fcis-1460 71 22 data datum NOUN fcis-1460 71 23 , , PUNCT fcis-1460 71 24 and and CCONJ fcis-1460 71 25 design design VERB fcis-1460 71 26 the the DET fcis-1460 71 27 corresponding correspond VERB fcis-1460 71 28 discrete discrete ADJ fcis-1460 71 29 pooling pool VERB fcis-1460 71 30 layer layer NOUN fcis-1460 71 31 , , PUNCT fcis-1460 71 32 and and CCONJ fcis-1460 71 33 then then ADV fcis-1460 71 34 combine combine VERB fcis-1460 71 35 the the DET fcis-1460 71 36 residual residual ADJ fcis-1460 71 37 structure structure NOUN fcis-1460 71 38 to to PART fcis-1460 71 39 design design VERB fcis-1460 71 40 discnet discnet NOUN fcis-1460 71 41 . . PUNCT fcis-1460 72 1 then then ADV fcis-1460 72 2 we we PRON fcis-1460 72 3 conduct conduct VERB fcis-1460 72 4 a a DET fcis-1460 72 5 comprehensive comprehensive ADJ fcis-1460 72 6 comparison comparison NOUN fcis-1460 72 7 with with ADP fcis-1460 72 8 some some DET fcis-1460 72 9 sota sota ADJ fcis-1460 72 10 algorithms algorithm NOUN fcis-1460 72 11 under under ADP fcis-1460 72 12 the the DET fcis-1460 72 13 cityscapes cityscape NOUN fcis-1460 72 14 datasets dataset NOUN fcis-1460 72 15 , , PUNCT fcis-1460 72 16 demonstrating demonstrate VERB fcis-1460 72 17 that that SCONJ fcis-1460 72 18 discnet discnet NOUN fcis-1460 72 19 performs perform VERB fcis-1460 72 20 well well ADV fcis-1460 72 21 in in ADP fcis-1460 72 22 both both PRON fcis-1460 72 23 accuracy accuracy NOUN fcis-1460 72 24 and and CCONJ fcis-1460 72 25 model model NOUN fcis-1460 72 26 size size NOUN fcis-1460 72 27 . . PUNCT fcis-1460 73 1 acknowledgements acknowledgement NOUN fcis-1460 73 2 this this DET fcis-1460 73 3 work work NOUN fcis-1460 73 4 is be AUX fcis-1460 73 5 supported support VERB fcis-1460 73 6 by by ADP fcis-1460 73 7 natural natural ADJ fcis-1460 73 8 science science NOUN fcis-1460 73 9 program program NOUN fcis-1460 73 10 of of ADP fcis-1460 73 11 guangdong guangdong PROPN fcis-1460 73 12 university university PROPN fcis-1460 73 13 of of ADP fcis-1460 73 14 science science NOUN fcis-1460 73 15 and and CCONJ fcis-1460 73 16 technology technology NOUN fcis-1460 73 17 under under ADP fcis-1460 73 18 the the DET fcis-1460 73 19 grant grant PROPN fcis-1460 73 20 no.gky-2021kyqnk-3 no.gky-2021kyqnk-3 PROPN fcis-1460 73 21 . . PUNCT fcis-1460 74 1 references reference NOUN fcis-1460 74 2 [ [ X fcis-1460 74 3 1 1 NUM fcis-1460 74 4 ] ] X fcis-1460 74 5 liu liu PROPN fcis-1460 74 6 s s PROPN fcis-1460 74 7 , , PUNCT fcis-1460 74 8 qi qi PROPN fcis-1460 74 9 l l NOUN fcis-1460 74 10 , , PUNCT fcis-1460 74 11 qin qin PROPN fcis-1460 74 12 h h PROPN fcis-1460 74 13 , , PUNCT fcis-1460 74 14 et et PROPN fcis-1460 74 15 al al PROPN fcis-1460 74 16 . . PROPN fcis-1460 74 17 path path PROPN fcis-1460 74 18 aggregation aggregation NOUN fcis-1460 74 19 network network NOUN fcis-1460 74 20 for for ADP fcis-1460 74 21 instance instance NOUN fcis-1460 74 22 segmentation[c]//proceedings segmentation[c]//proceeding NOUN fcis-1460 74 23 of of ADP fcis-1460 74 24 the the DET fcis-1460 74 25 ieee ieee NOUN fcis-1460 74 26 conference conference NOUN fcis-1460 74 27 on on ADP fcis-1460 74 28 computer computer NOUN fcis-1460 74 29 vision vision NOUN fcis-1460 74 30 and and CCONJ fcis-1460 74 31 pattern pattern NOUN fcis-1460 74 32 recognition recognition NOUN fcis-1460 74 33 . . PUNCT fcis-1460 75 1 2018 2018 NUM fcis-1460 75 2 : : PUNCT fcis-1460 75 3 8759 8759 NUM fcis-1460 75 4 - - SYM fcis-1460 75 5 8768 8768 NUM fcis-1460 75 6 . . PUNCT fcis-1460 76 1 [ [ X fcis-1460 76 2 2 2 NUM fcis-1460 76 3 ] ] SYM fcis-1460 76 4 li li PROPN fcis-1460 76 5 y y PROPN fcis-1460 76 6 , , PUNCT fcis-1460 76 7 qi qi PROPN fcis-1460 76 8 h h NOUN fcis-1460 76 9 , , PUNCT fcis-1460 76 10 dai dai PROPN fcis-1460 76 11 j j PROPN fcis-1460 76 12 , , PUNCT fcis-1460 76 13 et et PROPN fcis-1460 76 14 al al PROPN fcis-1460 76 15 . . PUNCT fcis-1460 76 16 fully fully ADV fcis-1460 76 17 convolutional convolutional ADJ fcis-1460 76 18 instance instance NOUN fcis-1460 76 19 - - PUNCT fcis-1460 76 20 aware aware ADJ fcis-1460 76 21 semantic semantic ADJ fcis-1460 76 22 segmentation[c]//proceedings segmentation[c]//proceeding NOUN fcis-1460 76 23 of of ADP fcis-1460 76 24 the the DET fcis-1460 76 25 ieee ieee NOUN fcis-1460 76 26 conference conference NOUN fcis-1460 76 27 on on ADP fcis-1460 76 28 computer computer NOUN fcis-1460 76 29 vision vision NOUN fcis-1460 76 30 and and CCONJ fcis-1460 76 31 pattern pattern NOUN fcis-1460 76 32 recognition recognition NOUN fcis-1460 76 33 . . PUNCT fcis-1460 77 1 2017 2017 NUM fcis-1460 77 2 : : PUNCT fcis-1460 77 3 2359 2359 NUM fcis-1460 77 4 - - SYM fcis-1460 77 5 2367 2367 NUM fcis-1460 77 6 . . PUNCT fcis-1460 78 1 [ [ X fcis-1460 78 2 3 3 NUM fcis-1460 78 3 ] ] PUNCT fcis-1460 78 4 xu xu PROPN fcis-1460 78 5 w w PROPN fcis-1460 78 6 , , PUNCT fcis-1460 78 7 wang wang PROPN fcis-1460 78 8 h h PROPN fcis-1460 78 9 , , PUNCT fcis-1460 78 10 qi qi PROPN fcis-1460 78 11 f f PROPN fcis-1460 78 12 , , PUNCT fcis-1460 78 13 et et PROPN fcis-1460 78 14 al al PROPN fcis-1460 78 15 . . PROPN fcis-1460 78 16 explicit explicit ADJ fcis-1460 78 17 shape shape NOUN fcis-1460 78 18 encoding encoding NOUN fcis-1460 78 19 for for ADP fcis-1460 78 20 realtime realtime ADJ fcis-1460 78 21 instance instance NOUN fcis-1460 78 22 segmentation[c]//proceedings segmentation[c]//proceeding NOUN fcis-1460 78 23 of of ADP fcis-1460 78 24 the the DET fcis-1460 78 25 ieee ieee NOUN fcis-1460 78 26 / / SYM fcis-1460 78 27 cvf cvf NOUN fcis-1460 78 28 international international ADJ fcis-1460 78 29 conference conference NOUN fcis-1460 78 30 on on ADP fcis-1460 78 31 computer computer NOUN fcis-1460 78 32 vision vision NOUN fcis-1460 78 33 . . PUNCT fcis-1460 79 1 2019 2019 NUM fcis-1460 79 2 : : PUNCT fcis-1460 79 3 51685177 51685177 NUM fcis-1460 79 4 . . PUNCT fcis-1460 80 1 [ [ X fcis-1460 80 2 4 4 NUM fcis-1460 80 3 ] ] SYM fcis-1460 80 4 yuan yuan NOUN fcis-1460 80 5 z z PROPN fcis-1460 80 6 w w PROPN fcis-1460 80 7 , , PUNCT fcis-1460 80 8 zhang zhang PROPN fcis-1460 80 9 j. j. PROPN fcis-1460 80 10 feature feature PROPN fcis-1460 80 11 extraction extraction NOUN fcis-1460 80 12 and and CCONJ fcis-1460 80 13 image image NOUN fcis-1460 80 14 retrieval retrieval NOUN fcis-1460 80 15 based base VERB fcis-1460 80 16 on on ADP fcis-1460 80 17 alexnet[c]//eighth alexnet[c]//eighth PROPN fcis-1460 80 18 international international ADJ fcis-1460 80 19 conference conference NOUN fcis-1460 80 20 on on ADP fcis-1460 80 21 digital digital ADJ fcis-1460 80 22 image image NOUN fcis-1460 80 23 processing processing NOUN fcis-1460 80 24 ( ( PUNCT fcis-1460 80 25 icdip icdip NOUN fcis-1460 80 26 2016 2016 NUM fcis-1460 80 27 ) ) PUNCT fcis-1460 80 28 . . PUNCT fcis-1460 81 1 spie spie NOUN fcis-1460 81 2 , , PUNCT fcis-1460 81 3 2016 2016 NUM fcis-1460 81 4 , , PUNCT fcis-1460 81 5 10033 10033 NUM fcis-1460 81 6 : : PUNCT fcis-1460 81 7 65 65 NUM fcis-1460 81 8 - - SYM fcis-1460 81 9 69 69 NUM fcis-1460 81 10 . . PUNCT fcis-1460 82 1 [ [ X fcis-1460 82 2 5 5 X fcis-1460 82 3 ] ] PUNCT fcis-1460 82 4 iandola iandola PROPN fcis-1460 82 5 f f PROPN fcis-1460 82 6 n n PROPN fcis-1460 82 7 , , PUNCT fcis-1460 82 8 han han PROPN fcis-1460 82 9 s s PROPN fcis-1460 82 10 , , PUNCT fcis-1460 82 11 moskewicz moskewicz PROPN fcis-1460 82 12 m m PROPN fcis-1460 82 13 w w PROPN fcis-1460 82 14 , , PUNCT fcis-1460 82 15 et et PROPN fcis-1460 82 16 al al PROPN fcis-1460 82 17 . . PUNCT fcis-1460 82 18 squeezenet squeezenet NOUN fcis-1460 82 19 : : PUNCT fcis-1460 82 20 alexnet alexnet ADJ fcis-1460 82 21 - - PUNCT fcis-1460 82 22 level level NOUN fcis-1460 82 23 accuracy accuracy NOUN fcis-1460 82 24 with with ADP fcis-1460 82 25 50x 50x NUM fcis-1460 82 26 fewer few ADJ fcis-1460 82 27 parameters parameter NOUN fcis-1460 82 28 and and CCONJ fcis-1460 82 29 < < X fcis-1460 82 30 0.5 0.5 NUM fcis-1460 82 31 mb mb PROPN fcis-1460 82 32 model model NOUN fcis-1460 82 33 size[j size[j PROPN fcis-1460 82 34 ] ] PUNCT fcis-1460 82 35 . . PUNCT fcis-1460 83 1 arxiv arxiv PROPN fcis-1460 83 2 preprint preprint PROPN fcis-1460 83 3 arxiv:1602.07360 arxiv:1602.07360 NOUN fcis-1460 83 4 , , PUNCT fcis-1460 83 5 2016 2016 NUM fcis-1460 83 6 . . PUNCT fcis-1460 84 1 [ [ X fcis-1460 84 2 6 6 X fcis-1460 84 3 ] ] PUNCT fcis-1460 84 4 szegedy szegedy NOUN fcis-1460 84 5 c c PROPN fcis-1460 84 6 , , PUNCT fcis-1460 84 7 vanhoucke vanhoucke NOUN fcis-1460 84 8 v v NOUN fcis-1460 84 9 , , PUNCT fcis-1460 84 10 ioffe ioffe PROPN fcis-1460 84 11 s s PART fcis-1460 84 12 , , PUNCT fcis-1460 84 13 et et PROPN fcis-1460 84 14 al al PROPN fcis-1460 84 15 . . PUNCT fcis-1460 85 1 rethinking rethink VERB fcis-1460 85 2 the the DET fcis-1460 85 3 inception inception ADJ fcis-1460 85 4 architecture architecture NOUN fcis-1460 85 5 for for ADP fcis-1460 85 6 computer computer NOUN fcis-1460 85 7 vision[j vision[j PROPN fcis-1460 85 8 ] ] PUNCT fcis-1460 85 9 . . PUNCT fcis-1460 86 1 ieee ieee PROPN fcis-1460 86 2 , , PUNCT fcis-1460 86 3 2016:2818 2016:2818 PROPN fcis-1460 86 4 - - SYM fcis-1460 86 5 2826 2826 NUM fcis-1460 86 6 . . PUNCT fcis-1460 87 1 [ [ X fcis-1460 87 2 7 7 X fcis-1460 87 3 ] ] X fcis-1460 87 4 howard howard PROPN fcis-1460 87 5 a a DET fcis-1460 87 6 g g PROPN fcis-1460 87 7 , , PUNCT fcis-1460 87 8 zhu zhu PROPN fcis-1460 87 9 m m PROPN fcis-1460 87 10 , , PUNCT fcis-1460 87 11 chen chen PROPN fcis-1460 87 12 b b PROPN fcis-1460 87 13 , , PUNCT fcis-1460 87 14 et et PROPN fcis-1460 87 15 al al PROPN fcis-1460 87 16 . . PROPN fcis-1460 87 17 mobilenets mobilenet NOUN fcis-1460 87 18 : : PUNCT fcis-1460 87 19 efficient efficient ADJ fcis-1460 87 20 convolutional convolutional ADJ fcis-1460 87 21 neural neural ADJ fcis-1460 87 22 networks network NOUN fcis-1460 87 23 for for ADP fcis-1460 87 24 mobile mobile ADJ fcis-1460 87 25 vision vision NOUN fcis-1460 87 26 applications[j applications[j PROPN fcis-1460 87 27 ] ] PUNCT fcis-1460 87 28 . . PUNCT fcis-1460 88 1 arxiv arxiv PROPN fcis-1460 88 2 preprint preprint PROPN fcis-1460 88 3 arxiv:1704.04861 arxiv:1704.04861 NOUN fcis-1460 88 4 , , PUNCT fcis-1460 88 5 2017 2017 NUM fcis-1460 88 6 . . PUNCT fcis-1460 89 1 [ [ X fcis-1460 89 2 8 8 NUM fcis-1460 89 3 ] ] X fcis-1460 89 4 peng peng PROPN fcis-1460 89 5 s s PROPN fcis-1460 89 6 , , PUNCT fcis-1460 89 7 jiang jiang PROPN fcis-1460 89 8 w w PROPN fcis-1460 89 9 , , PUNCT fcis-1460 89 10 pi pi PROPN fcis-1460 89 11 h h NOUN fcis-1460 89 12 , , PUNCT fcis-1460 89 13 et et PROPN fcis-1460 89 14 al al PROPN fcis-1460 89 15 . . PUNCT fcis-1460 90 1 deep deep ADJ fcis-1460 90 2 snake snake NOUN fcis-1460 90 3 for for ADP fcis-1460 90 4 real real ADJ fcis-1460 90 5 - - PUNCT fcis-1460 90 6 time time NOUN fcis-1460 90 7 instance instance NOUN fcis-1460 90 8 segmentation[c]//proceedings segmentation[c]//proceeding NOUN fcis-1460 90 9 of of ADP fcis-1460 90 10 the the DET fcis-1460 90 11 ieee ieee NOUN fcis-1460 90 12 / / SYM fcis-1460 90 13 cvf cvf NOUN fcis-1460 90 14 conference conference NOUN fcis-1460 90 15 on on ADP fcis-1460 90 16 computer computer NOUN fcis-1460 90 17 vision vision NOUN fcis-1460 90 18 and and CCONJ fcis-1460 90 19 pattern pattern NOUN fcis-1460 90 20 recognition recognition NOUN fcis-1460 90 21 . . PUNCT fcis-1460 91 1 2020 2020 NUM fcis-1460 91 2 : : PUNCT fcis-1460 91 3 8533 8533 NUM fcis-1460 91 4 - - SYM fcis-1460 91 5 8542 8542 NUM fcis-1460 91 6 . . PUNCT fcis-1460 92 1 47 47 NUM fcis-1460 93 1 [ [ X fcis-1460 93 2 9 9 NUM fcis-1460 93 3 ] ] X fcis-1460 93 4 zhao zhao PROPN fcis-1460 93 5 x x PROPN fcis-1460 93 6 , , PUNCT fcis-1460 93 7 vemulapalli vemulapalli PROPN fcis-1460 93 8 r r PROPN fcis-1460 93 9 , , PUNCT fcis-1460 93 10 mansfield mansfield PROPN fcis-1460 93 11 p p PROPN fcis-1460 93 12 a a PROPN fcis-1460 93 13 , , PUNCT fcis-1460 93 14 et et PROPN fcis-1460 93 15 al al PROPN fcis-1460 93 16 . . PROPN fcis-1460 93 17 contrastive contrastive ADJ fcis-1460 93 18 learning learning NOUN fcis-1460 93 19 for for ADP fcis-1460 93 20 label label NOUN fcis-1460 93 21 efficient efficient ADJ fcis-1460 93 22 semantic semantic ADJ fcis-1460 93 23 segmentation[c]//proceedings segmentation[c]//proceeding NOUN fcis-1460 93 24 of of ADP fcis-1460 93 25 the the DET fcis-1460 93 26 ieee ieee NOUN fcis-1460 93 27 / / SYM fcis-1460 93 28 cvf cvf NOUN fcis-1460 93 29 international international ADJ fcis-1460 93 30 conference conference NOUN fcis-1460 93 31 on on ADP fcis-1460 93 32 computer computer NOUN fcis-1460 93 33 vision vision NOUN fcis-1460 93 34 . . PUNCT fcis-1460 94 1 2021 2021 NUM fcis-1460 94 2 : : PUNCT fcis-1460 94 3 10623 10623 NUM fcis-1460 94 4 - - SYM fcis-1460 94 5 10633 10633 NUM fcis-1460 94 6 . . PUNCT fcis-1460 95 1 [ [ X fcis-1460 95 2 10 10 NUM fcis-1460 95 3 ] ] X fcis-1460 95 4 armato armato PROPN fcis-1460 95 5 s s PART fcis-1460 95 6 g g NOUN fcis-1460 95 7 , , PUNCT fcis-1460 95 8 roberts roberts PROPN fcis-1460 95 9 r r PROPN fcis-1460 95 10 y y PROPN fcis-1460 95 11 , , PUNCT fcis-1460 95 12 mcnitt mcnitt VERB fcis-1460 95 13 - - PUNCT fcis-1460 95 14 gray gray ADJ fcis-1460 95 15 m m NOUN fcis-1460 95 16 f f NOUN fcis-1460 95 17 , , PUNCT fcis-1460 95 18 et et PROPN fcis-1460 95 19 al al PROPN fcis-1460 95 20 . . PUNCT fcis-1460 96 1 the the DET fcis-1460 96 2 lung lung NOUN fcis-1460 96 3 image image NOUN fcis-1460 96 4 database database NOUN fcis-1460 96 5 consortium consortium PROPN fcis-1460 96 6 ( ( PUNCT fcis-1460 96 7 lidc lidc NOUN fcis-1460 96 8 ) ) PUNCT fcis-1460 96 9 and and CCONJ fcis-1460 96 10 image image NOUN fcis-1460 96 11 database database NOUN fcis-1460 96 12 resource resource NOUN fcis-1460 96 13 initiative initiative NOUN fcis-1460 96 14 ( ( PUNCT fcis-1460 96 15 idri idri ADJ fcis-1460 96 16 ): ): PUNCT fcis-1460 96 17 a a DET fcis-1460 96 18 completed complete VERB fcis-1460 96 19 reference reference NOUN fcis-1460 96 20 database database NOUN fcis-1460 96 21 of of ADP fcis-1460 96 22 lung lung NOUN fcis-1460 96 23 nodules nodule NOUN fcis-1460 96 24 on on ADP fcis-1460 96 25 ct ct PROPN fcis-1460 96 26 scans.[j scans.[j PROPN fcis-1460 96 27 ] ] PUNCT fcis-1460 96 28 . . PUNCT fcis-1460 97 1 academic academic ADJ fcis-1460 97 2 radiology radiology NOUN fcis-1460 97 3 , , PUNCT fcis-1460 97 4 2007 2007 NUM fcis-1460 97 5 , , PUNCT fcis-1460 97 6 14 14 NUM fcis-1460 97 7 ( ( PUNCT fcis-1460 97 8 12):1455 12):1455 NUM fcis-1460 97 9 - - SYM fcis-1460 97 10 1463 1463 NUM fcis-1460 97 11 . . PUNCT fcis-1460 98 1 [ [ X fcis-1460 98 2 11 11 NUM fcis-1460 98 3 ] ] PUNCT fcis-1460 98 4 jetley jetley PROPN fcis-1460 98 5 s s PROPN fcis-1460 98 6 , , PUNCT fcis-1460 98 7 sapienza sapienza PROPN fcis-1460 98 8 m m PROPN fcis-1460 98 9 , , PUNCT fcis-1460 98 10 golodetz golodetz VERB fcis-1460 98 11 s s PROPN fcis-1460 98 12 , , PUNCT fcis-1460 98 13 et et PROPN fcis-1460 98 14 al al PROPN fcis-1460 98 15 . . PUNCT fcis-1460 99 1 straight straight ADJ fcis-1460 99 2 to to ADP fcis-1460 99 3 shapes shape NOUN fcis-1460 99 4 : : PUNCT fcis-1460 99 5 realtime realtime NOUN fcis-1460 99 6 detection detection NOUN fcis-1460 99 7 of of ADP fcis-1460 99 8 encoded encode VERB fcis-1460 99 9 shapes[c]//proceedings shapes[c]//proceeding NOUN fcis-1460 99 10 of of ADP fcis-1460 99 11 the the DET fcis-1460 99 12 ieee ieee NOUN fcis-1460 99 13 conference conference NOUN fcis-1460 99 14 on on ADP fcis-1460 99 15 computer computer NOUN fcis-1460 99 16 vision vision NOUN fcis-1460 99 17 and and CCONJ fcis-1460 99 18 pattern pattern NOUN fcis-1460 99 19 recognition recognition NOUN fcis-1460 99 20 . . PUNCT fcis-1460 100 1 2017 2017 NUM fcis-1460 100 2 : : PUNCT fcis-1460 100 3 6550 6550 NUM fcis-1460 100 4 - - SYM fcis-1460 100 5 6559 6559 NUM fcis-1460 100 6 . . PUNCT fcis-1460 101 1 [ [ X fcis-1460 101 2 12 12 NUM fcis-1460 101 3 ] ] PUNCT fcis-1460 101 4 ze ze PROPN fcis-1460 101 5 yang yang PROPN fcis-1460 101 6 , , PUNCT fcis-1460 101 7 yinghao yinghao PROPN fcis-1460 101 8 xu xu PROPN fcis-1460 101 9 , , PUNCT fcis-1460 101 10 han han PROPN fcis-1460 101 11 xue xue PROPN fcis-1460 101 12 , , PUNCT fcis-1460 101 13 zheng zheng PROPN fcis-1460 101 14 zhang zhang PROPN fcis-1460 101 15 , , PUNCT fcis-1460 101 16 raquel raquel PROPN fcis-1460 101 17 urtasun urtasun PROPN fcis-1460 101 18 , , PUNCT fcis-1460 101 19 liwei liwei PROPN fcis-1460 101 20 wang wang PROPN fcis-1460 101 21 , , PUNCT fcis-1460 101 22 stephen stephen PROPN fcis-1460 101 23 lin lin PROPN fcis-1460 101 24 , , PUNCT fcis-1460 101 25 and and CCONJ fcis-1460 101 26 han han PROPN fcis-1460 101 27 hu hu PROPN fcis-1460 101 28 . . PUNCT fcis-1460 102 1 dense dense ADJ fcis-1460 102 2 reppoints reppoint NOUN fcis-1460 102 3 : : PUNCT fcis-1460 102 4 representing represent VERB fcis-1460 102 5 visual visual ADJ fcis-1460 102 6 objects object NOUN fcis-1460 102 7 with with ADP fcis-1460 102 8 dense dense ADJ fcis-1460 102 9 point point NOUN fcis-1460 102 10 sets set NOUN fcis-1460 102 11 . . PUNCT fcis-1460 103 1 arxiv arxiv PROPN fcis-1460 103 2 preprint preprint NOUN fcis-1460 103 3 arxiv:1912.11473 arxiv:1912.11473 NOUN fcis-1460 103 4 , , PUNCT fcis-1460 103 5 2019 2019 NUM fcis-1460 103 6 . . PUNCT fcis-1460 104 1 [ [ X fcis-1460 104 2 13 13 NUM fcis-1460 104 3 ] ] PUNCT fcis-1460 104 4 xingyi xingyi PROPN fcis-1460 104 5 zhou zhou PROPN fcis-1460 104 6 , , PUNCT fcis-1460 104 7 dequan dequan PROPN fcis-1460 104 8 wang wang PROPN fcis-1460 104 9 , , PUNCT fcis-1460 104 10 and and CCONJ fcis-1460 104 11 philipp philipp PROPN fcis-1460 104 12 kr kr PROPN fcis-1460 104 13 ähenb ähenb PROPN fcis-1460 104 14 ühl ühl PUNCT fcis-1460 104 15 . . PUNCT fcis-1460 105 1 objects object NOUN fcis-1460 105 2 as as ADP fcis-1460 105 3 points point NOUN fcis-1460 105 4 . . PUNCT fcis-1460 106 1 arxiv arxiv PROPN fcis-1460 106 2 preprint preprint NOUN fcis-1460 106 3 arxiv:1904.07850 arxiv:1904.07850 NOUN fcis-1460 106 4 , , PUNCT fcis-1460 106 5 2019 2019 NUM fcis-1460 106 6 . . PUNCT