id sid tid token lemma pos fcis-3736 1 1 frontiers frontier NOUN fcis-3736 1 2 in in ADP fcis-3736 1 3 computing computing NOUN fcis-3736 1 4 and and CCONJ fcis-3736 1 5 intelligent intelligent ADJ fcis-3736 1 6 systems system NOUN fcis-3736 1 7 issn issn VERB fcis-3736 1 8 : : PUNCT fcis-3736 1 9 2832 2832 NUM fcis-3736 1 10 - - SYM fcis-3736 1 11 6024 6024 NUM fcis-3736 1 12 | | NOUN fcis-3736 1 13 vol vol NOUN fcis-3736 1 14 . . PROPN fcis-3736 2 1 2 2 NUM fcis-3736 2 2 , , PUNCT fcis-3736 2 3 no no INTJ fcis-3736 2 4 . . NOUN fcis-3736 2 5 2 2 NUM fcis-3736 2 6 , , PUNCT fcis-3736 2 7 2022 2022 NUM fcis-3736 2 8 12 12 NUM fcis-3736 2 9 crowd crowd NOUN fcis-3736 2 10 counting counting NOUN fcis-3736 2 11 based base VERB fcis-3736 2 12 on on ADP fcis-3736 2 13 context context NOUN fcis-3736 2 14 - - PUNCT fcis-3736 2 15 aware aware ADJ fcis-3736 2 16 and and CCONJ fcis-3736 2 17 multi multi ADJ fcis-3736 2 18 scale scale NOUN fcis-3736 2 19 feature feature NOUN fcis-3736 2 20 fusion fusion NOUN fcis-3736 2 21 guoli guoli NOUN fcis-3736 2 22 zhang zhang PROPN fcis-3736 2 23 tiangong tiangong PROPN fcis-3736 2 24 university university PROPN fcis-3736 2 25 , , PUNCT fcis-3736 2 26 tianjin tianjin PROPN fcis-3736 2 27 300387 300387 NUM fcis-3736 2 28 , , PUNCT fcis-3736 2 29 china china PROPN fcis-3736 2 30 abstract abstract NOUN fcis-3736 2 31 : : PUNCT fcis-3736 2 32 crowd crowd NOUN fcis-3736 2 33 counting counting NOUN fcis-3736 2 34 plays play VERB fcis-3736 2 35 an an DET fcis-3736 2 36 important important ADJ fcis-3736 2 37 role role NOUN fcis-3736 2 38 in in ADP fcis-3736 2 39 public public ADJ fcis-3736 2 40 security security NOUN fcis-3736 2 41 . . PUNCT fcis-3736 3 1 estimating estimate VERB fcis-3736 3 2 the the DET fcis-3736 3 3 number number NOUN fcis-3736 3 4 of of ADP fcis-3736 3 5 people people NOUN fcis-3736 3 6 in in ADP fcis-3736 3 7 an an DET fcis-3736 3 8 image image NOUN fcis-3736 3 9 with with ADP fcis-3736 3 10 congested congested ADJ fcis-3736 3 11 crowd crowd NOUN fcis-3736 3 12 accurately accurately ADV fcis-3736 3 13 is be AUX fcis-3736 3 14 a a DET fcis-3736 3 15 challenging challenging ADJ fcis-3736 3 16 task task NOUN fcis-3736 3 17 . . PUNCT fcis-3736 4 1 the the DET fcis-3736 4 2 crowd crowd NOUN fcis-3736 4 3 counting counting NOUN fcis-3736 4 4 method method NOUN fcis-3736 4 5 based base VERB fcis-3736 4 6 on on ADP fcis-3736 4 7 fully fully ADV fcis-3736 4 8 convolutional convolutional ADJ fcis-3736 4 9 network network NOUN fcis-3736 4 10 can can AUX fcis-3736 4 11 perform perform VERB fcis-3736 4 12 well well ADV fcis-3736 4 13 in in ADP fcis-3736 4 14 crowd crowd NOUN fcis-3736 4 15 image image NOUN fcis-3736 4 16 with with ADP fcis-3736 4 17 complex complex ADJ fcis-3736 4 18 scene scene NOUN fcis-3736 4 19 . . PUNCT fcis-3736 5 1 in in ADP fcis-3736 5 2 this this DET fcis-3736 5 3 paper paper NOUN fcis-3736 5 4 , , PUNCT fcis-3736 5 5 to to PART fcis-3736 5 6 address address VERB fcis-3736 5 7 the the DET fcis-3736 5 8 counting counting NOUN fcis-3736 5 9 problems problem NOUN fcis-3736 5 10 of of ADP fcis-3736 5 11 occlusion，background occlusion，background ADJ fcis-3736 5 12 clutter clutter NOUN fcis-3736 5 13 and and CCONJ fcis-3736 5 14 perspective perspective ADJ fcis-3736 5 15 effect effect NOUN fcis-3736 5 16 , , PUNCT fcis-3736 5 17 we we PRON fcis-3736 5 18 proposed propose VERB fcis-3736 5 19 a a DET fcis-3736 5 20 simple simple ADJ fcis-3736 5 21 but but CCONJ fcis-3736 5 22 effective effective ADJ fcis-3736 5 23 method method NOUN fcis-3736 5 24 called call VERB fcis-3736 5 25 context context NOUN fcis-3736 5 26 - - PUNCT fcis-3736 5 27 aware aware ADJ fcis-3736 5 28 multi multi ADJ fcis-3736 5 29 scale scale NOUN fcis-3736 5 30 fusion fusion NOUN fcis-3736 5 31 network(cmf network(cmf NOUN fcis-3736 5 32 net).the net).the PRON fcis-3736 5 33 cmf cmf PROPN fcis-3736 5 34 net net PROPN fcis-3736 5 35 applied apply VERB fcis-3736 5 36 vgg vgg NOUN fcis-3736 5 37 network network NOUN fcis-3736 5 38 as as ADP fcis-3736 5 39 backbone backbone NOUN fcis-3736 5 40 to to PART fcis-3736 5 41 extract extract VERB fcis-3736 5 42 coarse coarse ADJ fcis-3736 5 43 features feature NOUN fcis-3736 5 44 . . PUNCT fcis-3736 6 1 then then ADV fcis-3736 6 2 , , PUNCT fcis-3736 6 3 three three NUM fcis-3736 6 4 context context NOUN fcis-3736 6 5 - - PUNCT fcis-3736 6 6 aware aware ADJ fcis-3736 6 7 multi multi ADJ fcis-3736 6 8 - - ADJ fcis-3736 6 9 scale scale ADJ fcis-3736 6 10 fusion fusion NOUN fcis-3736 6 11 modules module NOUN fcis-3736 6 12 ( ( PUNCT fcis-3736 6 13 cmfm cmfm ADV fcis-3736 6 14 ) ) PUNCT fcis-3736 6 15 are be AUX fcis-3736 6 16 adopted adopt VERB fcis-3736 6 17 . . PUNCT fcis-3736 7 1 each each DET fcis-3736 7 2 cmfm cmfm ADV fcis-3736 7 3 consist consist NOUN fcis-3736 7 4 of of ADP fcis-3736 7 5 multi multi ADJ fcis-3736 7 6 - - ADJ fcis-3736 7 7 scale scale ADJ fcis-3736 7 8 feature feature NOUN fcis-3736 7 9 extraction extraction NOUN fcis-3736 7 10 module module NOUN fcis-3736 7 11 ( ( PUNCT fcis-3736 7 12 mem mem ADJ fcis-3736 7 13 ) ) PUNCT fcis-3736 7 14 and and CCONJ fcis-3736 7 15 context context NOUN fcis-3736 7 16 - - PUNCT fcis-3736 7 17 aware aware ADJ fcis-3736 7 18 feature feature NOUN fcis-3736 7 19 extraction extraction NOUN fcis-3736 7 20 module module NOUN fcis-3736 7 21 ( ( PUNCT fcis-3736 7 22 cem cem NOUN fcis-3736 7 23 ) ) PUNCT fcis-3736 7 24 . . PUNCT fcis-3736 8 1 in in ADP fcis-3736 8 2 addition addition NOUN fcis-3736 8 3 , , PUNCT fcis-3736 8 4 we we PRON fcis-3736 8 5 propose propose VERB fcis-3736 8 6 adaptively adaptively ADV fcis-3736 8 7 dense dense ADJ fcis-3736 8 8 connection connection NOUN fcis-3736 8 9 to to PART fcis-3736 8 10 promoted promote VERB fcis-3736 8 11 information information NOUN fcis-3736 8 12 transmission transmission NOUN fcis-3736 8 13 in in ADP fcis-3736 8 14 the the DET fcis-3736 8 15 counting counting NOUN fcis-3736 8 16 network network NOUN fcis-3736 8 17 . . PUNCT fcis-3736 9 1 experiments experiment NOUN fcis-3736 9 2 on on ADP fcis-3736 9 3 four four NUM fcis-3736 9 4 datasets dataset NOUN fcis-3736 9 5 demonstrate demonstrate VERB fcis-3736 9 6 that that SCONJ fcis-3736 9 7 our our PRON fcis-3736 9 8 network network NOUN fcis-3736 9 9 achieves achieve VERB fcis-3736 9 10 competitive competitive ADJ fcis-3736 9 11 and and CCONJ fcis-3736 9 12 effective effective ADJ fcis-3736 9 13 results result NOUN fcis-3736 9 14 . . PUNCT fcis-3736 10 1 keywords keyword NOUN fcis-3736 10 2 : : PUNCT fcis-3736 10 3 crowd crowd NOUN fcis-3736 10 4 counting counting NOUN fcis-3736 10 5 ; ; PUNCT fcis-3736 10 6 deep deep ADJ fcis-3736 10 7 learning learning NOUN fcis-3736 10 8 ; ; PUNCT fcis-3736 10 9 computer computer NOUN fcis-3736 10 10 vision vision NOUN fcis-3736 10 11 ; ; PUNCT fcis-3736 10 12 attention attention NOUN fcis-3736 10 13 mechanism mechanism NOUN fcis-3736 10 14 ; ; PUNCT fcis-3736 10 15 feature feature NOUN fcis-3736 10 16 fusion fusion NOUN fcis-3736 10 17 . . PUNCT fcis-3736 11 1 1 1 X fcis-3736 11 2 . . X fcis-3736 11 3 introduction introduction NOUN fcis-3736 11 4 with with ADP fcis-3736 11 5 the the DET fcis-3736 11 6 development development NOUN fcis-3736 11 7 of of ADP fcis-3736 11 8 urbanization urbanization NOUN fcis-3736 11 9 , , PUNCT fcis-3736 11 10 a a DET fcis-3736 11 11 large large ADJ fcis-3736 11 12 number number NOUN fcis-3736 11 13 of of ADP fcis-3736 11 14 congested congested ADJ fcis-3736 11 15 crowd crowd NOUN fcis-3736 11 16 scenes scene NOUN fcis-3736 11 17 have have AUX fcis-3736 11 18 been be AUX fcis-3736 11 19 appeared appear VERB fcis-3736 11 20 in in ADP fcis-3736 11 21 large large ADJ fcis-3736 11 22 city city NOUN fcis-3736 11 23 . . PUNCT fcis-3736 12 1 and and CCONJ fcis-3736 12 2 it it PRON fcis-3736 12 3 hides hide VERB fcis-3736 12 4 many many ADJ fcis-3736 12 5 safety safety NOUN fcis-3736 12 6 hazards hazard NOUN fcis-3736 12 7 . . PUNCT fcis-3736 13 1 in in ADP fcis-3736 13 2 this this DET fcis-3736 13 3 year,29nd year,29nd PROPN fcis-3736 13 4 october october PROPN fcis-3736 13 5 2022 2022 NUM fcis-3736 13 6 , , PUNCT fcis-3736 13 7 over over ADP fcis-3736 13 8 150 150 NUM fcis-3736 13 9 people people NOUN fcis-3736 13 10 dead dead ADJ fcis-3736 13 11 after after ADP fcis-3736 13 12 stampede stampede NOUN fcis-3736 13 13 at at ADP fcis-3736 13 14 halloween halloween PROPN fcis-3736 13 15 event event NOUN fcis-3736 13 16 in in ADP fcis-3736 13 17 seoul seoul PROPN fcis-3736 13 18 , , PUNCT fcis-3736 13 19 south south PROPN fcis-3736 13 20 korea korea PROPN fcis-3736 13 21 . . PUNCT fcis-3736 14 1 therefore therefore ADV fcis-3736 14 2 , , PUNCT fcis-3736 14 3 researching research VERB fcis-3736 14 4 of of ADP fcis-3736 14 5 crowd crowd NOUN fcis-3736 14 6 counting counting NOUN fcis-3736 14 7 is be AUX fcis-3736 14 8 extremely extremely ADV fcis-3736 14 9 important important ADJ fcis-3736 14 10 in in ADP fcis-3736 14 11 intelligent intelligent ADJ fcis-3736 14 12 surveillance surveillance NOUN fcis-3736 14 13 , , PUNCT fcis-3736 14 14 disaster disaster NOUN fcis-3736 14 15 management management NOUN fcis-3736 14 16 and and CCONJ fcis-3736 14 17 social social ADJ fcis-3736 14 18 distance distance NOUN fcis-3736 14 19 monitoring monitoring NOUN fcis-3736 14 20 . . PUNCT fcis-3736 15 1 in in ADP fcis-3736 15 2 figure figure NOUN fcis-3736 15 3 1 1 NUM fcis-3736 15 4 , , PUNCT fcis-3736 15 5 we we PRON fcis-3736 15 6 shown show VERB fcis-3736 15 7 the the DET fcis-3736 15 8 crowd crowd NOUN fcis-3736 15 9 scenes scene NOUN fcis-3736 15 10 with with ADP fcis-3736 15 11 different different ADJ fcis-3736 15 12 density density NOUN fcis-3736 15 13 . . PUNCT fcis-3736 16 1 figure figure NOUN fcis-3736 16 2 1 1 NUM fcis-3736 16 3 . . PUNCT fcis-3736 17 1 crowd crowd NOUN fcis-3736 17 2 scenes scene NOUN fcis-3736 17 3 with with ADP fcis-3736 17 4 different different ADJ fcis-3736 17 5 density density NOUN fcis-3736 17 6 level level NOUN fcis-3736 17 7 traditional traditional ADJ fcis-3736 17 8 crowd crowd NOUN fcis-3736 17 9 counting counting NOUN fcis-3736 17 10 method method NOUN fcis-3736 17 11 based base VERB fcis-3736 17 12 on on ADP fcis-3736 17 13 object object NOUN fcis-3736 17 14 detection detection NOUN fcis-3736 17 15 [ [ X fcis-3736 17 16 1 1 X fcis-3736 17 17 ] ] PUNCT fcis-3736 17 18 is be AUX fcis-3736 17 19 not not PART fcis-3736 17 20 suitable suitable ADJ fcis-3736 17 21 for for ADP fcis-3736 17 22 highly highly ADV fcis-3736 17 23 congested congested ADJ fcis-3736 17 24 scene scene NOUN fcis-3736 17 25 . . PUNCT fcis-3736 18 1 inspired inspire VERB fcis-3736 18 2 by by ADP fcis-3736 18 3 the the DET fcis-3736 18 4 success success NOUN fcis-3736 18 5 of of ADP fcis-3736 18 6 deep deep ADJ fcis-3736 18 7 convolutional convolutional ADJ fcis-3736 18 8 networks network NOUN fcis-3736 18 9 , , PUNCT fcis-3736 18 10 varieties variety NOUN fcis-3736 18 11 of of ADP fcis-3736 18 12 crowd crowd NOUN fcis-3736 18 13 counting counting NOUN fcis-3736 18 14 method method NOUN fcis-3736 18 15 based base VERB fcis-3736 18 16 on on ADP fcis-3736 18 17 cnn cnn PROPN fcis-3736 18 18 have have AUX fcis-3736 18 19 been be AUX fcis-3736 18 20 proposed propose VERB fcis-3736 18 21 . . PUNCT fcis-3736 19 1 these these DET fcis-3736 19 2 networks network NOUN fcis-3736 19 3 usually usually ADV fcis-3736 19 4 adopt adopt VERB fcis-3736 19 5 an an DET fcis-3736 19 6 encoder encoder NOUN fcis-3736 19 7 - - PUNCT fcis-3736 19 8 decoder decoder NOUN fcis-3736 19 9 structure structure NOUN fcis-3736 19 10 , , PUNCT fcis-3736 19 11 and and CCONJ fcis-3736 19 12 generated generate VERB fcis-3736 19 13 the the DET fcis-3736 19 14 predicted predict VERB fcis-3736 19 15 density density NOUN fcis-3736 19 16 map map NOUN fcis-3736 19 17 . . PUNCT fcis-3736 20 1 due due ADP fcis-3736 20 2 to to ADP fcis-3736 20 3 perspective perspective ADJ fcis-3736 20 4 effect effect NOUN fcis-3736 20 5 of of ADP fcis-3736 20 6 different different ADJ fcis-3736 20 7 camera camera NOUN fcis-3736 20 8 viewpoint viewpoint NOUN fcis-3736 20 9 , , PUNCT fcis-3736 20 10 the the DET fcis-3736 20 11 scale scale NOUN fcis-3736 20 12 of of ADP fcis-3736 20 13 people people NOUN fcis-3736 20 14 in in ADP fcis-3736 20 15 the the DET fcis-3736 20 16 image image NOUN fcis-3736 20 17 varies vary VERB fcis-3736 20 18 dramatically dramatically ADV fcis-3736 20 19 . . PUNCT fcis-3736 21 1 and and CCONJ fcis-3736 21 2 counting count VERB fcis-3736 21 3 in in ADP fcis-3736 21 4 an an DET fcis-3736 21 5 image image NOUN fcis-3736 21 6 also also ADV fcis-3736 21 7 has have VERB fcis-3736 21 8 other other ADJ fcis-3736 21 9 problems problem NOUN fcis-3736 21 10 , , PUNCT fcis-3736 21 11 such such ADJ fcis-3736 21 12 as as ADP fcis-3736 21 13 occlusion occlusion NOUN fcis-3736 21 14 , , PUNCT fcis-3736 21 15 illumination illumination NOUN fcis-3736 21 16 changes change NOUN fcis-3736 21 17 , , PUNCT fcis-3736 21 18 background background NOUN fcis-3736 21 19 noises noise NOUN fcis-3736 21 20 and and CCONJ fcis-3736 21 21 clutter clutter NOUN fcis-3736 21 22 . . PUNCT fcis-3736 22 1 to to PART fcis-3736 22 2 solve solve VERB fcis-3736 22 3 these these DET fcis-3736 22 4 difficulties difficulty NOUN fcis-3736 22 5 , , PUNCT fcis-3736 22 6 mcnn mcnn NOUN fcis-3736 22 7 [ [ X fcis-3736 22 8 2 2 NUM fcis-3736 22 9 ] ] PUNCT fcis-3736 22 10 adopt adopt VERB fcis-3736 22 11 a a DET fcis-3736 22 12 multi multi ADJ fcis-3736 22 13 - - ADJ fcis-3736 22 14 column column ADJ fcis-3736 22 15 cnn cnn PROPN fcis-3736 22 16 network network NOUN fcis-3736 22 17 to to PART fcis-3736 22 18 capture capture VERB fcis-3736 22 19 scale scale NOUN fcis-3736 22 20 - - PUNCT fcis-3736 22 21 aware aware ADJ fcis-3736 22 22 features feature NOUN fcis-3736 22 23 with with ADP fcis-3736 22 24 different different ADJ fcis-3736 22 25 receptive receptive ADJ fcis-3736 22 26 fields field NOUN fcis-3736 22 27 . . PUNCT fcis-3736 23 1 cp cp PROPN fcis-3736 23 2 - - PROPN fcis-3736 23 3 cnn cnn PROPN fcis-3736 23 4 [ [ X fcis-3736 23 5 3 3 NUM fcis-3736 23 6 ] ] PUNCT fcis-3736 23 7 designed design VERB fcis-3736 23 8 a a DET fcis-3736 23 9 context context NOUN fcis-3736 23 10 - - PUNCT fcis-3736 23 11 aware aware ADJ fcis-3736 23 12 pyramid pyramid NOUN fcis-3736 23 13 network network NOUN fcis-3736 23 14 to to PART fcis-3736 23 15 combine combine VERB fcis-3736 23 16 the the DET fcis-3736 23 17 local local ADJ fcis-3736 23 18 and and CCONJ fcis-3736 23 19 global global ADJ fcis-3736 23 20 features feature NOUN fcis-3736 23 21 . . PUNCT fcis-3736 24 1 csr csr NOUN fcis-3736 24 2 - - ADJ fcis-3736 24 3 net[4 net[4 PRON fcis-3736 24 4 ] ] PUNCT fcis-3736 24 5 employed employ VERB fcis-3736 24 6 a a DET fcis-3736 24 7 deeper deep ADJ fcis-3736 24 8 single single ADJ fcis-3736 24 9 - - PUNCT fcis-3736 24 10 column column NOUN fcis-3736 24 11 network network NOUN fcis-3736 24 12 , , PUNCT fcis-3736 24 13 it it PRON fcis-3736 24 14 adopted adopt VERB fcis-3736 24 15 dilated dilate VERB fcis-3736 24 16 six six NUM fcis-3736 24 17 convolutional convolutional ADJ fcis-3736 24 18 layers layer NOUN fcis-3736 24 19 to to PART fcis-3736 24 20 further far ADV fcis-3736 24 21 enlarge enlarge VERB fcis-3736 24 22 receptive receptive ADJ fcis-3736 24 23 fields.[5 fields.[5 X fcis-3736 24 24 ] ] PUNCT fcis-3736 24 25 analyzed analyze VERB fcis-3736 24 26 the the DET fcis-3736 24 27 impact impact NOUN fcis-3736 24 28 of of ADP fcis-3736 24 29 the the DET fcis-3736 24 30 wrong wrong ADJ fcis-3736 24 31 predictions prediction NOUN fcis-3736 24 32 on on ADP fcis-3736 24 33 background background NOUN fcis-3736 24 34 regions region NOUN fcis-3736 24 35 in in ADP fcis-3736 24 36 crowd crowd NOUN fcis-3736 24 37 counting counting NOUN fcis-3736 24 38 . . PUNCT fcis-3736 25 1 in in ADP fcis-3736 25 2 summary summary NOUN fcis-3736 25 3 , , PUNCT fcis-3736 25 4 we we PRON fcis-3736 25 5 analyzed analyze VERB fcis-3736 25 6 that that SCONJ fcis-3736 25 7 the the DET fcis-3736 25 8 context context NOUN fcis-3736 25 9 information information NOUN fcis-3736 25 10 , , PUNCT fcis-3736 25 11 enlarging enlarge VERB fcis-3736 25 12 the the DET fcis-3736 25 13 receptive receptive ADJ fcis-3736 25 14 fields field NOUN fcis-3736 25 15 and and CCONJ fcis-3736 25 16 multi multi ADJ fcis-3736 25 17 - - ADJ fcis-3736 25 18 scale scale ADJ fcis-3736 25 19 feature feature NOUN fcis-3736 25 20 extraction extraction NOUN fcis-3736 25 21 are be AUX fcis-3736 25 22 the the DET fcis-3736 25 23 keys key NOUN fcis-3736 25 24 to to PART fcis-3736 25 25 solve solve VERB fcis-3736 25 26 the the DET fcis-3736 25 27 problems problem NOUN fcis-3736 25 28 of of ADP fcis-3736 25 29 crowd crowd NOUN fcis-3736 25 30 counting counting NOUN fcis-3736 25 31 . . PUNCT fcis-3736 26 1 in in ADP fcis-3736 26 2 this this DET fcis-3736 26 3 paper paper NOUN fcis-3736 26 4 , , PUNCT fcis-3736 26 5 the the DET fcis-3736 26 6 contributions contribution NOUN fcis-3736 26 7 of of ADP fcis-3736 26 8 our our PRON fcis-3736 26 9 work work NOUN fcis-3736 26 10 are be AUX fcis-3736 26 11 as as SCONJ fcis-3736 26 12 follows follow VERB fcis-3736 26 13 : : PUNCT fcis-3736 26 14 we we PRON fcis-3736 26 15 proposed propose VERB fcis-3736 26 16 a a DET fcis-3736 26 17 single single ADJ fcis-3736 26 18 - - PUNCT fcis-3736 26 19 column column NOUN fcis-3736 26 20 and and CCONJ fcis-3736 26 21 multi multi ADJ fcis-3736 26 22 - - ADJ fcis-3736 26 23 branch branch ADJ fcis-3736 26 24 structure structure NOUN fcis-3736 26 25 based base VERB fcis-3736 26 26 on on ADP fcis-3736 26 27 dilated dilated ADJ fcis-3736 26 28 convolutional convolutional ADJ fcis-3736 26 29 layers layer NOUN fcis-3736 26 30 with with ADP fcis-3736 26 31 different different ADJ fcis-3736 26 32 rates rate NOUN fcis-3736 26 33 to to PART fcis-3736 26 34 extract extract VERB fcis-3736 26 35 multi multi ADJ fcis-3736 26 36 - - ADJ fcis-3736 26 37 scale scale ADJ fcis-3736 26 38 features feature NOUN fcis-3736 26 39 . . PUNCT fcis-3736 27 1 it it PRON fcis-3736 27 2 can can AUX fcis-3736 27 3 solve solve VERB fcis-3736 27 4 the the DET fcis-3736 27 5 problem problem NOUN fcis-3736 27 6 of of ADP fcis-3736 27 7 perspective perspective ADJ fcis-3736 27 8 effect effect NOUN fcis-3736 27 9 . . PUNCT fcis-3736 28 1 and and CCONJ fcis-3736 28 2 we we PRON fcis-3736 28 3 proposed propose VERB fcis-3736 28 4 adaptively adaptively ADV fcis-3736 28 5 dense dense ADJ fcis-3736 28 6 connection connection NOUN fcis-3736 28 7 to to PART fcis-3736 28 8 promote promote VERB fcis-3736 28 9 the the DET fcis-3736 28 10 information information NOUN fcis-3736 28 11 transmission transmission NOUN fcis-3736 28 12 . . PUNCT fcis-3736 29 1 we we PRON fcis-3736 29 2 applied apply VERB fcis-3736 29 3 attention attention NOUN fcis-3736 29 4 mechanism mechanism NOUN fcis-3736 29 5 to to PART fcis-3736 29 6 capture capture VERB fcis-3736 29 7 context context NOUN fcis-3736 29 8 - - PUNCT fcis-3736 29 9 aware aware ADJ fcis-3736 29 10 features feature NOUN fcis-3736 29 11 . . PUNCT fcis-3736 30 1 this this DET fcis-3736 30 2 context context NOUN fcis-3736 30 3 information information NOUN fcis-3736 30 4 can can AUX fcis-3736 30 5 effectively effectively ADV fcis-3736 30 6 handle handle VERB fcis-3736 30 7 the the DET fcis-3736 30 8 large large ADJ fcis-3736 30 9 scenes scene NOUN fcis-3736 30 10 variations variation NOUN fcis-3736 30 11 among among ADP fcis-3736 30 12 the the DET fcis-3736 30 13 crowd crowd NOUN fcis-3736 30 14 images image NOUN fcis-3736 30 15 . . PUNCT fcis-3736 31 1 we we PRON fcis-3736 31 2 introduced introduce VERB fcis-3736 31 3 a a DET fcis-3736 31 4 novel novel ADJ fcis-3736 31 5 crowd crowd NOUN fcis-3736 31 6 counting count VERB fcis-3736 31 7 network network NOUN fcis-3736 31 8 ( ( PUNCT fcis-3736 31 9 cmf cmf PROPN fcis-3736 31 10 net net PROPN fcis-3736 31 11 ) ) PUNCT fcis-3736 31 12 by by ADP fcis-3736 31 13 end end NOUN fcis-3736 31 14 - - PUNCT fcis-3736 31 15 to to ADP fcis-3736 31 16 - - PUNCT fcis-3736 31 17 end end NOUN fcis-3736 31 18 structure structure NOUN fcis-3736 31 19 . . PUNCT fcis-3736 32 1 extensive extensive ADJ fcis-3736 32 2 experiments experiment NOUN fcis-3736 32 3 on on ADP fcis-3736 32 4 four four NUM fcis-3736 32 5 datasets dataset NOUN fcis-3736 32 6 demonstrate demonstrate VERB fcis-3736 32 7 that that SCONJ fcis-3736 32 8 our our PRON fcis-3736 32 9 network network NOUN fcis-3736 32 10 is be AUX fcis-3736 32 11 effective effective ADJ fcis-3736 32 12 and and CCONJ fcis-3736 32 13 competitive competitive ADJ fcis-3736 32 14 to to PART fcis-3736 32 15 handle handle VERB fcis-3736 32 16 the the DET fcis-3736 32 17 problem problem NOUN fcis-3736 32 18 of of ADP fcis-3736 32 19 perspective perspective ADJ fcis-3736 32 20 effect effect NOUN fcis-3736 32 21 . . PUNCT fcis-3736 33 1 2 2 X fcis-3736 33 2 . . NUM fcis-3736 33 3 proposed propose VERB fcis-3736 33 4 method method NOUN fcis-3736 33 5 in in ADP fcis-3736 33 6 this this DET fcis-3736 33 7 section section NOUN fcis-3736 33 8 , , PUNCT fcis-3736 33 9 we we PRON fcis-3736 33 10 firstly firstly ADV fcis-3736 33 11 described describe VERB fcis-3736 33 12 the the DET fcis-3736 33 13 overall overall ADJ fcis-3736 33 14 network network NOUN fcis-3736 33 15 of of ADP fcis-3736 33 16 our our PRON fcis-3736 33 17 cmf cmf PROPN fcis-3736 33 18 net net NOUN fcis-3736 33 19 in in ADP fcis-3736 33 20 figure figure NOUN fcis-3736 33 21 2 2 NUM fcis-3736 33 22 . . PUNCT fcis-3736 34 1 and and CCONJ fcis-3736 34 2 then then ADV fcis-3736 34 3 we we PRON fcis-3736 34 4 have have VERB fcis-3736 34 5 to to PART fcis-3736 34 6 introduce introduce VERB fcis-3736 34 7 the the DET fcis-3736 34 8 detailed detailed ADJ fcis-3736 34 9 information information NOUN fcis-3736 34 10 about about ADP fcis-3736 34 11 multi multi ADJ fcis-3736 34 12 - - ADJ fcis-3736 34 13 scale scale ADJ fcis-3736 34 14 feature feature NOUN fcis-3736 34 15 extraction extraction NOUN fcis-3736 34 16 module module NOUN fcis-3736 34 17 ( ( PUNCT fcis-3736 34 18 msm msm NOUN fcis-3736 34 19 ) ) PUNCT fcis-3736 34 20 , , PUNCT fcis-3736 34 21 context context NOUN fcis-3736 34 22 feature feature NOUN fcis-3736 34 23 extraction extraction NOUN fcis-3736 34 24 module module NOUN fcis-3736 34 25 ( ( PUNCT fcis-3736 34 26 cfm cfm NOUN fcis-3736 34 27 ) ) PUNCT fcis-3736 34 28 and and CCONJ fcis-3736 34 29 adaptively adaptively ADV fcis-3736 34 30 dense dense ADJ fcis-3736 34 31 connection connection NOUN fcis-3736 34 32 . . PUNCT fcis-3736 35 1 2.1 2.1 NUM fcis-3736 35 2 . . PUNCT fcis-3736 35 3 overview overview NOUN fcis-3736 35 4 crowd crowd NOUN fcis-3736 35 5 counting counting NOUN fcis-3736 35 6 task task NOUN fcis-3736 35 7 is be AUX fcis-3736 35 8 regarded regard VERB fcis-3736 35 9 as as ADP fcis-3736 35 10 a a DET fcis-3736 35 11 density density NOUN fcis-3736 35 12 map map NOUN fcis-3736 35 13 regression regression NOUN fcis-3736 35 14 problem problem NOUN fcis-3736 35 15 from from ADP fcis-3736 35 16 an an DET fcis-3736 35 17 image image NOUN fcis-3736 35 18 . . PUNCT fcis-3736 36 1 we we PRON fcis-3736 36 2 adopted adopt VERB fcis-3736 36 3 the the DET fcis-3736 36 4 first first ADJ fcis-3736 36 5 10 10 NUM fcis-3736 36 6 layers layer NOUN fcis-3736 36 7 of of ADP fcis-3736 36 8 vgg16[6 vgg16[6 PROPN fcis-3736 36 9 ] ] X fcis-3736 36 10 networks network NOUN fcis-3736 36 11 as as ADP fcis-3736 36 12 the the DET fcis-3736 36 13 backbone backbone NOUN fcis-3736 36 14 of of ADP fcis-3736 36 15 the the DET fcis-3736 36 16 cmf cmf PROPN fcis-3736 36 17 net net PROPN fcis-3736 36 18 because because SCONJ fcis-3736 36 19 of of ADP fcis-3736 36 20 it it PRON fcis-3736 36 21 can can AUX fcis-3736 36 22 balance balance VERB fcis-3736 36 23 feature feature NOUN fcis-3736 36 24 extraction extraction NOUN fcis-3736 36 25 efficiency efficiency NOUN fcis-3736 36 26 and and CCONJ fcis-3736 36 27 computational computational ADJ fcis-3736 36 28 complexity complexity NOUN fcis-3736 36 29 . . PUNCT fcis-3736 37 1 due due ADP fcis-3736 37 2 to to PART fcis-3736 37 3 avoid avoid VERB fcis-3736 37 4 overfitting overfitte VERB fcis-3736 37 5 and and CCONJ fcis-3736 37 6 the the DET fcis-3736 37 7 number number NOUN fcis-3736 37 8 of of ADP fcis-3736 37 9 dataset dataset ADJ fcis-3736 37 10 imagers imager NOUN fcis-3736 37 11 is be AUX fcis-3736 37 12 not not PART fcis-3736 37 13 sufficient sufficient ADJ fcis-3736 37 14 , , PUNCT fcis-3736 37 15 we we PRON fcis-3736 37 16 adopted adopt VERB fcis-3736 37 17 transfer transfer NOUN fcis-3736 37 18 learning learning NOUN fcis-3736 37 19 on on ADP fcis-3736 37 20 imagenet imagenet ADJ fcis-3736 37 21 dataset dataset NOUN fcis-3736 37 22 of of ADP fcis-3736 37 23 classification classification NOUN fcis-3736 37 24 task task NOUN fcis-3736 37 25 to to ADP fcis-3736 37 26 pre pre ADJ fcis-3736 37 27 - - ADJ fcis-3736 37 28 train train ADJ fcis-3736 37 29 vgg16 vgg16 NOUN fcis-3736 37 30 network network NOUN fcis-3736 37 31 . . PUNCT fcis-3736 38 1 the the DET fcis-3736 38 2 backbone backbone NOUN fcis-3736 38 3 ’s ’s PART fcis-3736 38 4 output output NOUN fcis-3736 38 5 resolution resolution NOUN fcis-3736 38 6 is be AUX fcis-3736 38 7 1/8 1/8 NUM fcis-3736 38 8 size size NOUN fcis-3736 38 9 of of ADP fcis-3736 38 10 original original ADJ fcis-3736 38 11 input input NOUN fcis-3736 38 12 image image NOUN fcis-3736 38 13 . . PUNCT fcis-3736 39 1 the the DET fcis-3736 39 2 output output NOUN fcis-3736 39 3 feature feature NOUN fcis-3736 39 4 of of ADP fcis-3736 39 5 image image NOUN fcis-3736 39 6 i i PRON fcis-3736 39 7 are be AUX fcis-3736 39 8 generated generate VERB fcis-3736 39 9 by by ADP fcis-3736 39 10 : : PUNCT fcis-3736 39 11 f f PROPN fcis-3736 39 12 = = PROPN fcis-3736 39 13 𝑣 𝑣 PROPN fcis-3736 39 14 β„± β„± PROPN fcis-3736 39 15 ( ( PUNCT fcis-3736 39 16 𝐼)𝑣𝑔𝑔 𝐼)𝑣𝑔𝑔 X fcis-3736 39 17 ( ( PUNCT fcis-3736 39 18 1 1 NUM fcis-3736 39 19 ) ) PUNCT fcis-3736 39 20 then then ADV fcis-3736 39 21 , , PUNCT fcis-3736 39 22 we we PRON fcis-3736 39 23 applied apply VERB fcis-3736 39 24 context context NOUN fcis-3736 39 25 - - PUNCT fcis-3736 39 26 aware aware ADJ fcis-3736 39 27 multi multi ADJ fcis-3736 39 28 - - ADJ fcis-3736 39 29 scale scale ADJ fcis-3736 39 30 fusion fusion NOUN fcis-3736 39 31 13 13 NUM fcis-3736 39 32 modules module NOUN fcis-3736 39 33 ( ( PUNCT fcis-3736 39 34 cmfm cmfm ADV fcis-3736 39 35 ) ) PUNCT fcis-3736 39 36 to to PART fcis-3736 39 37 capture capture VERB fcis-3736 39 38 multi multi ADJ fcis-3736 39 39 - - ADJ fcis-3736 39 40 scale scale ADJ fcis-3736 39 41 features feature NOUN fcis-3736 39 42 and and CCONJ fcis-3736 39 43 context context NOUN fcis-3736 39 44 information information NOUN fcis-3736 39 45 . . PUNCT fcis-3736 40 1 subsequently subsequently ADV fcis-3736 40 2 , , PUNCT fcis-3736 40 3 we we PRON fcis-3736 40 4 enhanced enhance VERB fcis-3736 40 5 the the DET fcis-3736 40 6 resolution resolution NOUN fcis-3736 40 7 of of ADP fcis-3736 40 8 the the DET fcis-3736 40 9 feature feature NOUN fcis-3736 40 10 map map NOUN fcis-3736 40 11 by by ADP fcis-3736 40 12 up up ADP fcis-3736 40 13 sampling sample VERB fcis-3736 40 14 with with ADP fcis-3736 40 15 the the DET fcis-3736 40 16 factor factor NOUN fcis-3736 40 17 of of ADP fcis-3736 40 18 2 2 NUM fcis-3736 40 19 . . PUNCT fcis-3736 41 1 we we PRON fcis-3736 41 2 repeated repeat VERB fcis-3736 41 3 this this DET fcis-3736 41 4 procedure procedure NOUN fcis-3736 41 5 for for ADP fcis-3736 41 6 3 3 NUM fcis-3736 41 7 times time NOUN fcis-3736 41 8 and and CCONJ fcis-3736 41 9 use use VERB fcis-3736 41 10 adaptively adaptively ADV fcis-3736 41 11 dense dense ADJ fcis-3736 41 12 connections connection NOUN fcis-3736 41 13 to to PART fcis-3736 41 14 increase increase VERB fcis-3736 41 15 information information NOUN fcis-3736 41 16 transmission transmission NOUN fcis-3736 41 17 . . PUNCT fcis-3736 42 1 finally finally ADV fcis-3736 42 2 , , PUNCT fcis-3736 42 3 two two NUM fcis-3736 42 4 dilated dilated ADJ fcis-3736 42 5 and and CCONJ fcis-3736 42 6 one one NUM fcis-3736 42 7 kernel kernel NOUN fcis-3736 42 8 size size NOUN fcis-3736 42 9 of of ADP fcis-3736 42 10 1 1 NUM fcis-3736 42 11 * * SYM fcis-3736 42 12 1 1 NUM fcis-3736 42 13 convolutional convolutional ADJ fcis-3736 42 14 layer layer NOUN fcis-3736 42 15 as as ADP fcis-3736 42 16 back back ADJ fcis-3736 42 17 end end NOUN fcis-3736 42 18 to to PART fcis-3736 42 19 generate generate VERB fcis-3736 42 20 a a DET fcis-3736 42 21 density density NOUN fcis-3736 42 22 map map NOUN fcis-3736 42 23 of of ADP fcis-3736 42 24 input input NOUN fcis-3736 42 25 image image NOUN fcis-3736 42 26 . . PUNCT fcis-3736 43 1 2.2 2.2 NUM fcis-3736 43 2 . . PUNCT fcis-3736 44 1 context context NOUN fcis-3736 44 2 - - PUNCT fcis-3736 44 3 aware aware ADJ fcis-3736 44 4 multi multi ADJ fcis-3736 44 5 - - ADJ fcis-3736 44 6 scale scale ADJ fcis-3736 44 7 fusion fusion NOUN fcis-3736 44 8 module module NOUN fcis-3736 44 9 due due ADP fcis-3736 44 10 to to ADP fcis-3736 44 11 perspective perspective ADJ fcis-3736 44 12 effect effect NOUN fcis-3736 44 13 , , PUNCT fcis-3736 44 14 the the DET fcis-3736 44 15 people people NOUN fcis-3736 44 16 scale scale VERB fcis-3736 44 17 in in ADP fcis-3736 44 18 one one NUM fcis-3736 44 19 image image NOUN fcis-3736 44 20 change change NOUN fcis-3736 44 21 dramatically dramatically ADV fcis-3736 44 22 . . PUNCT fcis-3736 45 1 and and CCONJ fcis-3736 45 2 the the DET fcis-3736 45 3 scenes scene NOUN fcis-3736 45 4 in in ADP fcis-3736 45 5 training training NOUN fcis-3736 45 6 set set NOUN fcis-3736 45 7 differ differ VERB fcis-3736 45 8 from from ADP fcis-3736 45 9 one one NUM fcis-3736 45 10 another another DET fcis-3736 45 11 . . PUNCT fcis-3736 46 1 [ [ X fcis-3736 46 2 7 7 X fcis-3736 46 3 ] ] PUNCT fcis-3736 46 4 proved prove VERB fcis-3736 46 5 that that SCONJ fcis-3736 46 6 context context PROPN fcis-3736 46 7 information information NOUN fcis-3736 46 8 is be AUX fcis-3736 46 9 helpful helpful ADJ fcis-3736 46 10 for for ADP fcis-3736 46 11 crowd crowd NOUN fcis-3736 46 12 counting counting NOUN fcis-3736 46 13 task task NOUN fcis-3736 46 14 . . PUNCT fcis-3736 47 1 therefore therefore ADV fcis-3736 47 2 , , PUNCT fcis-3736 47 3 fusing fuse VERB fcis-3736 47 4 the the DET fcis-3736 47 5 multi multi ADJ fcis-3736 47 6 scale scale NOUN fcis-3736 47 7 features feature NOUN fcis-3736 47 8 and and CCONJ fcis-3736 47 9 context context NOUN fcis-3736 47 10 information information NOUN fcis-3736 47 11 is be AUX fcis-3736 47 12 the the DET fcis-3736 47 13 main main ADJ fcis-3736 47 14 solution solution NOUN fcis-3736 47 15 of of ADP fcis-3736 47 16 the the DET fcis-3736 47 17 problem problem NOUN fcis-3736 47 18 . . PUNCT fcis-3736 48 1 we we PRON fcis-3736 48 2 proposed propose VERB fcis-3736 48 3 context context NOUN fcis-3736 48 4 - - PUNCT fcis-3736 48 5 aware aware ADJ fcis-3736 48 6 multi multi ADJ fcis-3736 48 7 - - ADJ fcis-3736 48 8 scale scale ADJ fcis-3736 48 9 fusion fusion NOUN fcis-3736 48 10 module module NOUN fcis-3736 48 11 ( ( PUNCT fcis-3736 48 12 cmfm cmfm PROPN fcis-3736 48 13 ) ) PUNCT fcis-3736 48 14 , , PUNCT fcis-3736 48 15 it it PRON fcis-3736 48 16 consist consist VERB fcis-3736 48 17 of of ADP fcis-3736 48 18 multi multi ADJ fcis-3736 48 19 - - ADJ fcis-3736 48 20 scale scale ADJ fcis-3736 48 21 feature feature NOUN fcis-3736 48 22 extraction extraction NOUN fcis-3736 48 23 module module NOUN fcis-3736 48 24 ( ( PUNCT fcis-3736 48 25 mem mem ADJ fcis-3736 48 26 ) ) PUNCT fcis-3736 48 27 and and CCONJ fcis-3736 48 28 context context NOUN fcis-3736 48 29 - - PUNCT fcis-3736 48 30 aware aware ADJ fcis-3736 48 31 feature feature NOUN fcis-3736 48 32 extraction extraction NOUN fcis-3736 48 33 module module NOUN fcis-3736 48 34 ( ( PUNCT fcis-3736 48 35 cem cem NOUN fcis-3736 48 36 ) ) PUNCT fcis-3736 48 37 . . PUNCT fcis-3736 49 1 at at ADP fcis-3736 49 2 last last ADV fcis-3736 49 3 , , PUNCT fcis-3736 49 4 we we PRON fcis-3736 49 5 fused fuse VERB fcis-3736 49 6 these these DET fcis-3736 49 7 features feature NOUN fcis-3736 49 8 with with ADP fcis-3736 49 9 residual residual ADJ fcis-3736 49 10 connection connection NOUN fcis-3736 49 11 , , PUNCT fcis-3736 49 12 element element NOUN fcis-3736 49 13 - - ADJ fcis-3736 49 14 wise wise ADJ fcis-3736 49 15 product product NOUN fcis-3736 49 16 and and CCONJ fcis-3736 49 17 addition addition NOUN fcis-3736 49 18 . . PUNCT fcis-3736 50 1 the the DET fcis-3736 50 2 description description NOUN fcis-3736 50 3 of of ADP fcis-3736 50 4 cmfm cmfm PROPN fcis-3736 50 5 in in ADP fcis-3736 50 6 figure figure NOUN fcis-3736 50 7 3 3 NUM fcis-3736 50 8 . . NOUN fcis-3736 50 9 2.2.1 2.2.1 NUM fcis-3736 50 10 . . PUNCT fcis-3736 51 1 multi multi ADJ fcis-3736 51 2 - - ADJ fcis-3736 51 3 scale scale ADJ fcis-3736 51 4 feature feature NOUN fcis-3736 51 5 extraction extraction NOUN fcis-3736 51 6 module module NOUN fcis-3736 51 7 ( ( PUNCT fcis-3736 51 8 mem mem ADJ fcis-3736 51 9 ) ) PUNCT fcis-3736 51 10 inspired inspire VERB fcis-3736 51 11 by by ADP fcis-3736 51 12 aspp aspp NOUN fcis-3736 51 13 [ [ X fcis-3736 51 14 8 8 NUM fcis-3736 51 15 ] ] PUNCT fcis-3736 51 16 , , PUNCT fcis-3736 51 17 we we PRON fcis-3736 51 18 designed design VERB fcis-3736 51 19 a a DET fcis-3736 51 20 multi multi ADJ fcis-3736 51 21 branch branch NOUN fcis-3736 51 22 pyramid pyramid NOUN fcis-3736 51 23 structure structure NOUN fcis-3736 51 24 with with ADP fcis-3736 51 25 aurous aurous ADJ fcis-3736 51 26 convolution convolution NOUN fcis-3736 51 27 layers layer NOUN fcis-3736 51 28 to to PART fcis-3736 51 29 capture capture VERB fcis-3736 51 30 multi multi ADJ fcis-3736 51 31 scale scale NOUN fcis-3736 51 32 features feature NOUN fcis-3736 51 33 . . PUNCT fcis-3736 52 1 each each DET fcis-3736 52 2 module module NOUN fcis-3736 52 3 has have VERB fcis-3736 52 4 four four NUM fcis-3736 52 5 parallel parallel ADJ fcis-3736 52 6 dilated dilate VERB fcis-3736 52 7 convolutional convolutional ADJ fcis-3736 52 8 layers layer NOUN fcis-3736 52 9 with with ADP fcis-3736 52 10 different different ADJ fcis-3736 52 11 dilation dilation NOUN fcis-3736 52 12 rates rate NOUN fcis-3736 52 13 . . PUNCT fcis-3736 53 1 the the DET fcis-3736 53 2 output output NOUN fcis-3736 53 3 feature feature NOUN fcis-3736 53 4 of of ADP fcis-3736 53 5 mem mem NOUN fcis-3736 53 6 contains contain VERB fcis-3736 53 7 multi multi ADJ fcis-3736 53 8 scale scale NOUN fcis-3736 53 9 information information NOUN fcis-3736 53 10 of of ADP fcis-3736 53 11 crowd crowd NOUN fcis-3736 53 12 image image NOUN fcis-3736 53 13 . . PUNCT fcis-3736 54 1 then then ADV fcis-3736 54 2 , , PUNCT fcis-3736 54 3 to to PART fcis-3736 54 4 balance balance VERB fcis-3736 54 5 the the DET fcis-3736 54 6 counting counting NOUN fcis-3736 54 7 efficiency efficiency NOUN fcis-3736 54 8 and and CCONJ fcis-3736 54 9 the the DET fcis-3736 54 10 number number NOUN fcis-3736 54 11 of of ADP fcis-3736 54 12 computational computational ADJ fcis-3736 54 13 parameters parameter NOUN fcis-3736 54 14 , , PUNCT fcis-3736 54 15 we we PRON fcis-3736 54 16 adopted adopt VERB fcis-3736 54 17 the the DET fcis-3736 54 18 way way NOUN fcis-3736 54 19 of of ADP fcis-3736 54 20 cascade cascade NOUN fcis-3736 54 21 to to PART fcis-3736 54 22 fuse fuse NOUN fcis-3736 54 23 pyramid pyramid NOUN fcis-3736 54 24 features feature NOUN fcis-3736 54 25 . . PUNCT fcis-3736 55 1 2.2.2 2.2.2 X fcis-3736 55 2 . . PUNCT fcis-3736 55 3 context context NOUN fcis-3736 55 4 - - PUNCT fcis-3736 55 5 aware aware ADJ fcis-3736 55 6 feature feature NOUN fcis-3736 55 7 extraction extraction NOUN fcis-3736 55 8 module module NOUN fcis-3736 55 9 ( ( PUNCT fcis-3736 55 10 cem cem NOUN fcis-3736 55 11 ) ) PUNCT fcis-3736 55 12 we we PRON fcis-3736 55 13 utilized utilize VERB fcis-3736 55 14 attention attention NOUN fcis-3736 55 15 mechanism mechanism NOUN fcis-3736 55 16 to to PART fcis-3736 55 17 extract extract VERB fcis-3736 55 18 the the DET fcis-3736 55 19 contextaware contextaware NOUN fcis-3736 55 20 information information NOUN fcis-3736 55 21 . . PUNCT fcis-3736 56 1 this this DET fcis-3736 56 2 pixel pixel ADJ fcis-3736 56 3 - - ADJ fcis-3736 56 4 wise wise ADJ fcis-3736 56 5 context context NOUN fcis-3736 56 6 information information NOUN fcis-3736 56 7 based base VERB fcis-3736 56 8 on on ADP fcis-3736 56 9 attention attention NOUN fcis-3736 56 10 can can AUX fcis-3736 56 11 highlight highlight VERB fcis-3736 56 12 the the DET fcis-3736 56 13 foreground foreground NOUN fcis-3736 56 14 information information NOUN fcis-3736 56 15 of of ADP fcis-3736 56 16 people people NOUN fcis-3736 56 17 ’s ’s PART fcis-3736 56 18 head head NOUN fcis-3736 56 19 and and CCONJ fcis-3736 56 20 suppress suppress VERB fcis-3736 56 21 the the DET fcis-3736 56 22 noise noise NOUN fcis-3736 56 23 of of ADP fcis-3736 56 24 background background NOUN fcis-3736 56 25 region region NOUN fcis-3736 56 26 . . PUNCT fcis-3736 57 1 making make VERB fcis-3736 57 2 the the DET fcis-3736 57 3 network network NOUN fcis-3736 57 4 pay pay VERB fcis-3736 57 5 more more ADJ fcis-3736 57 6 attention attention NOUN fcis-3736 57 7 to to ADP fcis-3736 57 8 the the DET fcis-3736 57 9 crowd crowd NOUN fcis-3736 57 10 region region NOUN fcis-3736 57 11 . . PUNCT fcis-3736 58 1 the the DET fcis-3736 58 2 cem cem NOUN fcis-3736 58 3 consist consist NOUN fcis-3736 58 4 of of ADP fcis-3736 58 5 channel channel NOUN fcis-3736 58 6 attention attention NOUN fcis-3736 58 7 branch branch NOUN fcis-3736 58 8 and and CCONJ fcis-3736 58 9 space space NOUN fcis-3736 58 10 attention attention NOUN fcis-3736 58 11 branch branch NOUN fcis-3736 58 12 . . PUNCT fcis-3736 59 1 we we PRON fcis-3736 59 2 devided devide VERB fcis-3736 59 3 the the DET fcis-3736 59 4 cbam cbam NOUN fcis-3736 59 5 figure figure NOUN fcis-3736 59 6 2 2 NUM fcis-3736 59 7 . . PUNCT fcis-3736 60 1 the the DET fcis-3736 60 2 architecture architecture NOUN fcis-3736 60 3 of of ADP fcis-3736 60 4 our our PRON fcis-3736 60 5 proposed propose VERB fcis-3736 60 6 cmf cmf PROPN fcis-3736 60 7 net net PROPN fcis-3736 60 8 . . PUNCT fcis-3736 61 1 the the DET fcis-3736 61 2 convolution convolution NOUN fcis-3736 61 3 layers layer NOUN fcis-3736 61 4 are be AUX fcis-3736 61 5 designed design VERB fcis-3736 61 6 as as ADP fcis-3736 61 7 conv conv ADJ fcis-3736 61 8 ( ( PUNCT fcis-3736 61 9 input input NOUN fcis-3736 61 10 channels channel NOUN fcis-3736 61 11 , , PUNCT fcis-3736 61 12 output output NOUN fcis-3736 61 13 channels channel NOUN fcis-3736 61 14 , , PUNCT fcis-3736 61 15 kernel kernel NOUN fcis-3736 61 16 size size NOUN fcis-3736 61 17 , , PUNCT fcis-3736 61 18 padding padding NOUN fcis-3736 61 19 , , PUNCT fcis-3736 61 20 dilation dilation NOUN fcis-3736 61 21 rate rate NOUN fcis-3736 61 22 ) ) PUNCT fcis-3736 61 23 network network NOUN fcis-3736 61 24 [ [ X fcis-3736 61 25 9 9 NUM fcis-3736 61 26 ] ] PUNCT fcis-3736 61 27 to to PART fcis-3736 61 28 get get VERB fcis-3736 61 29 the the DET fcis-3736 61 30 two two NUM fcis-3736 61 31 branches branch NOUN fcis-3736 61 32 . . PUNCT fcis-3736 62 1 and and CCONJ fcis-3736 62 2 combined combine VERB fcis-3736 62 3 them they PRON fcis-3736 62 4 in in ADP fcis-3736 62 5 parallel parallel ADJ fcis-3736 62 6 way way NOUN fcis-3736 62 7 . . PUNCT fcis-3736 63 1 the the DET fcis-3736 63 2 attention attention NOUN fcis-3736 63 3 masks mask NOUN fcis-3736 63 4 generated generate VERB fcis-3736 63 5 by by ADP fcis-3736 63 6 cem cem NOUN fcis-3736 63 7 produced produce VERB fcis-3736 63 8 the the DET fcis-3736 63 9 multi multi ADJ fcis-3736 63 10 - - ADJ fcis-3736 63 11 scale scale ADJ fcis-3736 63 12 feature feature NOUN fcis-3736 63 13 from from ADP fcis-3736 63 14 the the DET fcis-3736 63 15 mem mem NOUN fcis-3736 63 16 . . PUNCT fcis-3736 64 1 at at ADP fcis-3736 64 2 last last ADV fcis-3736 64 3 , , PUNCT fcis-3736 64 4 we we PRON fcis-3736 64 5 applied apply VERB fcis-3736 64 6 residual residual ADJ fcis-3736 64 7 connection connection NOUN fcis-3736 64 8 and and CCONJ fcis-3736 64 9 element element NOUN fcis-3736 64 10 - - ADJ fcis-3736 64 11 wise wise ADJ fcis-3736 64 12 addition addition NOUN fcis-3736 64 13 to to AUX fcis-3736 64 14 fuse fuse VERB fcis-3736 64 15 context context NOUN fcis-3736 64 16 - - PUNCT fcis-3736 64 17 aware aware ADJ fcis-3736 64 18 features feature NOUN fcis-3736 64 19 and and CCONJ fcis-3736 64 20 original original ADJ fcis-3736 64 21 feature feature NOUN fcis-3736 64 22 map map NOUN fcis-3736 64 23 . . PUNCT fcis-3736 65 1 the the DET fcis-3736 65 2 procedure procedure NOUN fcis-3736 65 3 can can AUX fcis-3736 65 4 be be AUX fcis-3736 65 5 formulated formulate VERB fcis-3736 65 6 as as SCONJ fcis-3736 65 7 follows follow VERB fcis-3736 65 8 : : PUNCT fcis-3736 65 9 𝑀𝑐(𝐹𝑖𝑛 𝑀𝑐(𝐹𝑖𝑛 NOUN fcis-3736 65 10 ) ) PUNCT fcis-3736 65 11 = = SYM fcis-3736 65 12 𝜎 𝜎 PROPN fcis-3736 65 13 ( ( PUNCT fcis-3736 65 14 𝑀𝐿𝑃(π΄π‘£π‘”π‘ƒπ‘œπ‘œπ‘™(𝐹𝑖𝑛 𝑀𝐿𝑃(π΄π‘£π‘”π‘ƒπ‘œπ‘œπ‘™(𝐹𝑖𝑛 NOUN fcis-3736 65 15 ) ) PUNCT fcis-3736 65 16 ) ) PUNCT fcis-3736 66 1 + + CCONJ fcis-3736 66 2 𝑀𝐿𝑃(π‘€π‘Žπ‘₯π‘ƒπ‘œπ‘œπ‘™(𝐹𝑖𝑛 𝑀𝐿𝑃(π‘€π‘Žπ‘₯π‘ƒπ‘œπ‘œπ‘™(𝐹𝑖𝑛 NOUN fcis-3736 66 3 ) ) PUNCT fcis-3736 66 4 ) ) PUNCT fcis-3736 66 5 ) ) PUNCT fcis-3736 67 1 ( ( PUNCT fcis-3736 67 2 2 2 X fcis-3736 67 3 ) ) PUNCT fcis-3736 67 4 𝑀𝑠(𝐹𝑖𝑛 𝑀𝑠(𝐹𝑖𝑛 NOUN fcis-3736 67 5 ) ) PUNCT fcis-3736 67 6 = = SYM fcis-3736 67 7 𝜎(𝑓7Γ—7([π΄π‘£π‘”π‘ƒπ‘œπ‘œπ‘™(𝐹𝑖𝑛 𝜎(𝑓7Γ—7([π΄π‘£π‘”π‘ƒπ‘œπ‘œπ‘™(𝐹𝑖𝑛 NOUN fcis-3736 67 8 ) ) PUNCT fcis-3736 67 9 ; ; PUNCT fcis-3736 68 1 π‘€π‘Žπ‘₯π‘ƒπ‘œπ‘œπ‘™(𝐹𝑖𝑛 π‘€π‘Žπ‘₯π‘ƒπ‘œπ‘œπ‘™(𝐹𝑖𝑛 PROPN fcis-3736 68 2 ) ) PUNCT fcis-3736 68 3 ] ] PUNCT fcis-3736 68 4 ) ) PUNCT fcis-3736 68 5 ) ) PUNCT fcis-3736 69 1 ( ( PUNCT fcis-3736 69 2 3 3 X fcis-3736 69 3 ) ) PUNCT fcis-3736 69 4 fout fout NOUN fcis-3736 70 1 = = SYM fcis-3736 70 2 mc mc PROPN fcis-3736 70 3 Γ— Γ— PROPN fcis-3736 70 4 fms fms PROPN fcis-3736 71 1 + + CCONJ fcis-3736 71 2 ms ms PROPN fcis-3736 71 3 Γ— Γ— PROPN fcis-3736 71 4 fms fms PROPN fcis-3736 71 5 + + CCONJ fcis-3736 71 6 fin fin NOUN fcis-3736 71 7 ( ( PUNCT fcis-3736 71 8 4 4 NUM fcis-3736 71 9 ) ) PUNCT fcis-3736 71 10 where where SCONJ fcis-3736 71 11 𝑀𝑐(𝐹𝑖𝑛 𝑀𝑐(𝐹𝑖𝑛 NOUN fcis-3736 71 12 ) ) PUNCT fcis-3736 71 13 represents represent VERB fcis-3736 71 14 the the DET fcis-3736 71 15 channel channel NOUN fcis-3736 71 16 - - PUNCT fcis-3736 71 17 wise wise ADJ fcis-3736 71 18 attention attention NOUN fcis-3736 71 19 weights weight NOUN fcis-3736 71 20 generated generate VERB fcis-3736 71 21 by by ADP fcis-3736 71 22 channel channel NOUN fcis-3736 71 23 attention attention NOUN fcis-3736 71 24 branch branch NOUN fcis-3736 71 25 of of ADP fcis-3736 71 26 cem cem NOUN fcis-3736 71 27 and and CCONJ fcis-3736 71 28 𝑀𝑠(𝐹𝑖𝑛 𝑀𝑠(𝐹𝑖𝑛 NOUN fcis-3736 71 29 ) ) PUNCT fcis-3736 71 30 represents represent VERB fcis-3736 71 31 the the DET fcis-3736 71 32 spatial spatial ADJ fcis-3736 71 33 attention attention NOUN fcis-3736 71 34 weights weight NOUN fcis-3736 71 35 . . PUNCT fcis-3736 72 1 mlp mlp NOUN fcis-3736 72 2 is be AUX fcis-3736 72 3 the the DET fcis-3736 72 4 abbreviation abbreviation NOUN fcis-3736 72 5 of of ADP fcis-3736 72 6 multi multi ADJ fcis-3736 72 7 - - ADJ fcis-3736 72 8 layer layer ADJ fcis-3736 72 9 perceptron perceptron PROPN fcis-3736 72 10 . . PUNCT fcis-3736 73 1 Οƒ Οƒ PROPN fcis-3736 73 2 denotes denote NOUN fcis-3736 73 3 sigmoid sigmoid NOUN fcis-3736 73 4 function function NOUN fcis-3736 73 5 . . PUNCT fcis-3736 74 1 𝑓7Γ—7 𝑓7Γ—7 PROPN fcis-3736 74 2 represent represent VERB fcis-3736 74 3 the the DET fcis-3736 74 4 convolutional convolutional ADJ fcis-3736 74 5 layer layer NOUN fcis-3736 74 6 with with ADP fcis-3736 74 7 the the DET fcis-3736 74 8 kernel kernel NOUN fcis-3736 74 9 size size NOUN fcis-3736 74 10 of of ADP fcis-3736 74 11 7Γ—7.𝐹𝑖𝑛 7Γ—7.𝐹𝑖𝑛 PROPN fcis-3736 74 12 , , PUNCT fcis-3736 74 13 πΉπ‘šπ‘ ,πΉπ‘œπ‘’π‘‘denotes πΉπ‘šπ‘ ,πΉπ‘œπ‘’π‘‘denotes PROPN fcis-3736 74 14 the the DET fcis-3736 74 15 input input NOUN fcis-3736 74 16 feature feature NOUN fcis-3736 74 17 map map NOUN fcis-3736 74 18 , , PUNCT fcis-3736 74 19 multi multi ADJ fcis-3736 74 20 - - ADJ fcis-3736 74 21 scale scale ADJ fcis-3736 74 22 feature feature NOUN fcis-3736 74 23 and and CCONJ fcis-3736 74 24 the the DET fcis-3736 74 25 output output NOUN fcis-3736 74 26 feature feature NOUN fcis-3736 74 27 map map NOUN fcis-3736 74 28 of of ADP fcis-3736 74 29 the the DET fcis-3736 74 30 whole whole ADJ fcis-3736 74 31 cmfm cmfm ADV fcis-3736 74 32 respectively respectively ADV fcis-3736 74 33 . . PUNCT fcis-3736 75 1 2.3 2.3 NUM fcis-3736 75 2 . . PUNCT fcis-3736 76 1 adaptively adaptively ADV fcis-3736 76 2 dense dense ADJ fcis-3736 76 3 connection connection NOUN fcis-3736 76 4 the the DET fcis-3736 76 5 decoder decoder NOUN fcis-3736 76 6 of of ADP fcis-3736 76 7 cmf cmf PROPN fcis-3736 76 8 net net PROPN fcis-3736 76 9 adopted adopt VERB fcis-3736 76 10 gradually gradually ADV fcis-3736 76 11 up up ADP fcis-3736 76 12 sample sample NOUN fcis-3736 76 13 strategy strategy NOUN fcis-3736 76 14 . . PUNCT fcis-3736 77 1 thus thus ADV fcis-3736 77 2 , , PUNCT fcis-3736 77 3 common common ADJ fcis-3736 77 4 dense dense ADJ fcis-3736 77 5 connection connection NOUN fcis-3736 77 6 [ [ X fcis-3736 77 7 10 10 NUM fcis-3736 77 8 ] ] PUNCT fcis-3736 77 9 is be AUX fcis-3736 77 10 not not PART fcis-3736 77 11 suitable suitable ADJ fcis-3736 77 12 for for ADP fcis-3736 77 13 our our PRON fcis-3736 77 14 network network NOUN fcis-3736 77 15 . . PUNCT fcis-3736 78 1 we we PRON fcis-3736 78 2 designed design VERB fcis-3736 78 3 a a DET fcis-3736 78 4 refine refine ADJ fcis-3736 78 5 block block NOUN fcis-3736 78 6 to to PART fcis-3736 78 7 up up ADP fcis-3736 78 8 sample sample NOUN fcis-3736 78 9 and and CCONJ fcis-3736 78 10 refine refine VERB fcis-3736 78 11 the the DET fcis-3736 78 12 low low ADJ fcis-3736 78 13 - - PUNCT fcis-3736 78 14 resolution resolution NOUN fcis-3736 78 15 feature feature NOUN fcis-3736 78 16 map map NOUN fcis-3736 78 17 . . PUNCT fcis-3736 79 1 then then ADV fcis-3736 79 2 , , PUNCT fcis-3736 79 3 we we PRON fcis-3736 79 4 combined combine VERB fcis-3736 79 5 the the DET fcis-3736 79 6 refine refine ADJ fcis-3736 79 7 block block NOUN fcis-3736 79 8 and and CCONJ fcis-3736 79 9 dense dense ADJ fcis-3736 79 10 connection connection NOUN fcis-3736 79 11 as as ADP fcis-3736 79 12 adaptively adaptively ADV fcis-3736 79 13 dense dense ADJ fcis-3736 79 14 connection connection NOUN fcis-3736 79 15 . . PUNCT fcis-3736 80 1 the the DET fcis-3736 80 2 detailed detailed ADJ fcis-3736 80 3 description description NOUN fcis-3736 80 4 of of ADP fcis-3736 80 5 adaptively adaptively ADV fcis-3736 80 6 dense dense ADJ fcis-3736 80 7 connection connection NOUN fcis-3736 80 8 is be AUX fcis-3736 80 9 in in ADP fcis-3736 80 10 figure figure NOUN fcis-3736 80 11 4 4 NUM fcis-3736 80 12 . . PUNCT fcis-3736 81 1 it it PRON fcis-3736 81 2 is be AUX fcis-3736 81 3 helpful helpful ADJ fcis-3736 81 4 for for ADP fcis-3736 81 5 the the DET fcis-3736 81 6 back back ADJ fcis-3736 81 7 propagation propagation NOUN fcis-3736 81 8 during during ADP fcis-3736 81 9 the the DET fcis-3736 81 10 training training NOUN fcis-3736 81 11 and and CCONJ fcis-3736 81 12 enhance enhance VERB fcis-3736 81 13 the the DET fcis-3736 81 14 information information NOUN fcis-3736 81 15 transmission transmission NOUN fcis-3736 81 16 , , PUNCT fcis-3736 81 17 thereby thereby ADV fcis-3736 81 18 we we PRON fcis-3736 81 19 can can AUX fcis-3736 81 20 apply apply VERB fcis-3736 81 21 deeper deep ADJ fcis-3736 81 22 network network NOUN fcis-3736 81 23 . . PUNCT fcis-3736 82 1 2.4 2.4 NUM fcis-3736 82 2 . . PUNCT fcis-3736 83 1 loss loss NOUN fcis-3736 83 2 function function NOUN fcis-3736 83 3 we we PRON fcis-3736 83 4 use use VERB fcis-3736 83 5 euclidean euclidean ADJ fcis-3736 83 6 distance distance NOUN fcis-3736 83 7 loss loss NOUN fcis-3736 83 8 as as SCONJ fcis-3736 83 9 the the DET fcis-3736 83 10 loss loss NOUN fcis-3736 83 11 function function NOUN fcis-3736 83 12 of of ADP fcis-3736 83 13 our our PRON fcis-3736 83 14 network network NOUN fcis-3736 83 15 to to PART fcis-3736 83 16 measure measure VERB fcis-3736 83 17 the the DET fcis-3736 83 18 pixel pixel ADJ fcis-3736 83 19 - - ADJ fcis-3736 83 20 wise wise ADJ fcis-3736 83 21 difference difference NOUN fcis-3736 83 22 between between ADP fcis-3736 83 23 the the DET fcis-3736 83 24 predicted predict VERB fcis-3736 83 25 density density NOUN fcis-3736 83 26 map map NOUN fcis-3736 83 27 and and CCONJ fcis-3736 83 28 the the DET fcis-3736 83 29 ground ground NOUN fcis-3736 83 30 truth truth NOUN fcis-3736 83 31 density density NOUN fcis-3736 83 32 map map NOUN fcis-3736 83 33 , , PUNCT fcis-3736 83 34 the the DET fcis-3736 83 35 description description NOUN fcis-3736 83 36 of of ADP fcis-3736 83 37 loss loss NOUN fcis-3736 83 38 function function NOUN fcis-3736 83 39 are be AUX fcis-3736 83 40 as as SCONJ fcis-3736 83 41 follows follow VERB fcis-3736 83 42 : : PUNCT fcis-3736 83 43 l(xi l(xi ADJ fcis-3736 83 44 , , PUNCT fcis-3736 83 45 ΞΈ ΞΈ NOUN fcis-3736 83 46 ) ) PUNCT fcis-3736 83 47 = = SYM fcis-3736 84 1 1 1 NUM fcis-3736 84 2 2n 2n NUM fcis-3736 84 3 βˆ‘ βˆ‘ PUNCT fcis-3736 84 4 ||f(xi ||f(xi PROPN fcis-3736 84 5 , , PUNCT fcis-3736 84 6 ΞΈ ΞΈ NOUN fcis-3736 84 7 ) ) PUNCT fcis-3736 84 8 βˆ’ βˆ’ PROPN fcis-3736 84 9 di||2 di||2 NOUN fcis-3736 84 10 2n 2n NUM fcis-3736 84 11 i=1 i=1 X fcis-3736 84 12 ( ( PUNCT fcis-3736 84 13 5 5 NUM fcis-3736 84 14 ) ) PUNCT fcis-3736 84 15 where where SCONJ fcis-3736 84 16 n n PRON fcis-3736 84 17 represents represent VERB fcis-3736 84 18 the the DET fcis-3736 84 19 number number NOUN fcis-3736 84 20 of of ADP fcis-3736 84 21 training training NOUN fcis-3736 84 22 images image NOUN fcis-3736 84 23 . . PUNCT fcis-3736 85 1 xi xi PROPN fcis-3736 85 2 is be AUX fcis-3736 85 3 the the DET fcis-3736 85 4 input input NOUN fcis-3736 85 5 image image NOUN fcis-3736 85 6 . . PUNCT fcis-3736 86 1 ΞΈ ΞΈ NOUN fcis-3736 86 2 are be AUX fcis-3736 86 3 the the DET fcis-3736 86 4 parameters parameter NOUN fcis-3736 86 5 of of ADP fcis-3736 86 6 our our PRON fcis-3736 86 7 network network NOUN fcis-3736 86 8 ? ? PUNCT fcis-3736 87 1 f(xi f(xi PROPN fcis-3736 87 2 , , PUNCT fcis-3736 87 3 ΞΈ ΞΈ NOUN fcis-3736 87 4 ) ) PUNCT fcis-3736 87 5 represents represent VERB fcis-3736 87 6 the the DET fcis-3736 87 7 predicted predict VERB fcis-3736 87 8 density density NOUN fcis-3736 87 9 map map NOUN fcis-3736 87 10 and and CCONJ fcis-3736 87 11 the the DET fcis-3736 87 12 di di NOUN fcis-3736 87 13 represents represent VERB fcis-3736 87 14 the the DET fcis-3736 87 15 ground ground NOUN fcis-3736 87 16 truth truth NOUN fcis-3736 87 17 density density NOUN fcis-3736 87 18 map map NOUN fcis-3736 87 19 . . PUNCT fcis-3736 88 1 figure figure VERB fcis-3736 88 2 3 3 NUM fcis-3736 88 3 . . PUNCT fcis-3736 89 1 the the DET fcis-3736 89 2 architecture architecture NOUN fcis-3736 89 3 of of ADP fcis-3736 89 4 cmfm cmfm PROPN fcis-3736 89 5 . . PUNCT fcis-3736 90 1 the the DET fcis-3736 90 2 cam cam NOUN fcis-3736 90 3 and and CCONJ fcis-3736 90 4 sam sam PROPN fcis-3736 90 5 represent represent VERB fcis-3736 90 6 the the DET fcis-3736 90 7 channel channel NOUN fcis-3736 90 8 attention attention NOUN fcis-3736 90 9 module module NOUN fcis-3736 90 10 ( ( PUNCT fcis-3736 90 11 eq.2 eq.2 NOUN fcis-3736 90 12 . . PUNCT fcis-3736 90 13 ) ) PUNCT fcis-3736 91 1 and and CCONJ fcis-3736 91 2 the the DET fcis-3736 91 3 spatial spatial ADJ fcis-3736 91 4 attention attention NOUN fcis-3736 91 5 module module NOUN fcis-3736 91 6 ( ( PUNCT fcis-3736 91 7 eq.3 eq.3 PROPN fcis-3736 91 8 . . PUNCT fcis-3736 91 9 ) ) PUNCT fcis-3736 91 10 . . PUNCT fcis-3736 92 1 figure figure VERB fcis-3736 92 2 4 4 NUM fcis-3736 92 3 . . PUNCT fcis-3736 93 1 this this PRON fcis-3736 93 2 is be AUX fcis-3736 93 3 the the DET fcis-3736 93 4 description description NOUN fcis-3736 93 5 of of ADP fcis-3736 93 6 our our PRON fcis-3736 93 7 adaptively adaptively ADV fcis-3736 93 8 dense dense ADJ fcis-3736 93 9 connection connection NOUN fcis-3736 93 10 3 3 NUM fcis-3736 93 11 . . PUNCT fcis-3736 94 1 experiments experiment NOUN fcis-3736 94 2 in in ADP fcis-3736 94 3 this this DET fcis-3736 94 4 section section NOUN fcis-3736 94 5 , , PUNCT fcis-3736 94 6 firstly firstly ADV fcis-3736 94 7 we we PRON fcis-3736 94 8 introduced introduce VERB fcis-3736 94 9 the the DET fcis-3736 94 10 evaluation evaluation NOUN fcis-3736 94 11 metrics metric NOUN fcis-3736 94 12 . . PUNCT fcis-3736 95 1 then then ADV fcis-3736 95 2 , , PUNCT fcis-3736 95 3 we we PRON fcis-3736 95 4 described describe VERB fcis-3736 95 5 our our PRON fcis-3736 95 6 implementation implementation NOUN fcis-3736 95 7 details detail NOUN fcis-3736 95 8 . . PUNCT fcis-3736 96 1 at at ADP fcis-3736 96 2 last last ADV fcis-3736 96 3 , , PUNCT fcis-3736 96 4 we we PRON fcis-3736 96 5 compared compare VERB fcis-3736 96 6 our our PRON fcis-3736 96 7 network network NOUN fcis-3736 96 8 to to ADP fcis-3736 96 9 other other ADJ fcis-3736 96 10 functions function NOUN fcis-3736 96 11 3.1 3.1 NUM fcis-3736 96 12 . . PUNCT fcis-3736 96 13 evaluation evaluation NOUN fcis-3736 96 14 metrics metric NOUN fcis-3736 96 15 consistent consistent ADJ fcis-3736 96 16 with with ADP fcis-3736 96 17 the the DET fcis-3736 96 18 mainstream mainstream NOUN fcis-3736 96 19 methods method NOUN fcis-3736 96 20 , , PUNCT fcis-3736 96 21 we we PRON fcis-3736 96 22 adopted adopt VERB fcis-3736 96 23 mean mean VERB fcis-3736 96 24 absolute absolute ADJ fcis-3736 96 25 error error NOUN fcis-3736 96 26 ( ( PUNCT fcis-3736 96 27 mae mae PROPN fcis-3736 96 28 ) ) PUNCT fcis-3736 96 29 and and CCONJ fcis-3736 96 30 mean mean VERB fcis-3736 96 31 square square ADJ fcis-3736 96 32 error error NOUN fcis-3736 96 33 ( ( PUNCT fcis-3736 96 34 mse mse NOUN fcis-3736 96 35 ) ) PUNCT fcis-3736 96 36 to to PART fcis-3736 96 37 evaluate evaluate VERB fcis-3736 96 38 the the DET fcis-3736 96 39 performance performance NOUN fcis-3736 96 40 of of ADP fcis-3736 96 41 our our PRON fcis-3736 96 42 network network NOUN fcis-3736 96 43 , , PUNCT fcis-3736 96 44 which which PRON fcis-3736 96 45 are be AUX fcis-3736 96 46 defined define VERB fcis-3736 96 47 as as ADP fcis-3736 96 48 : : PUNCT fcis-3736 96 49 mae mae PROPN fcis-3736 96 50 = = PROPN fcis-3736 96 51 1 1 NUM fcis-3736 96 52 n n ADP fcis-3736 96 53 βˆ‘ βˆ‘ ADV fcis-3736 96 54 |ci |ci X fcis-3736 96 55 pre pre X fcis-3736 96 56 βˆ’ βˆ’ PROPN fcis-3736 96 57 ci ci PROPN fcis-3736 97 1 gt gt INTJ fcis-3736 98 1 | | ADV fcis-3736 98 2 n n PROPN fcis-3736 98 3 i=1 i=1 PROPN fcis-3736 98 4 14 14 NUM fcis-3736 98 5 and and CCONJ fcis-3736 98 6 mse mse NOUN fcis-3736 98 7 = = NOUN fcis-3736 98 8 √ √ PROPN fcis-3736 98 9 1 1 NUM fcis-3736 98 10 𝑁 𝑁 PROPN fcis-3736 98 11 βˆ‘(𝐢𝑖 βˆ‘(𝐢𝑖 PROPN fcis-3736 98 12 π‘π‘Ÿπ‘’ π‘π‘Ÿπ‘’ NOUN fcis-3736 98 13 βˆ’ βˆ’ PROPN fcis-3736 98 14 𝐢𝑖 𝐢𝑖 PROPN fcis-3736 98 15 𝑔𝑑 𝑔𝑑 ADP fcis-3736 98 16 ) ) PUNCT fcis-3736 98 17 2 2 NUM fcis-3736 99 1 𝑁 𝑁 PROPN fcis-3736 99 2 𝑖=1 𝑖=1 PROPN fcis-3736 99 3 where where SCONJ fcis-3736 99 4 n n PRON fcis-3736 99 5 is be AUX fcis-3736 99 6 the the DET fcis-3736 99 7 number number NOUN fcis-3736 99 8 of of ADP fcis-3736 99 9 images image NOUN fcis-3736 99 10 . . PUNCT fcis-3736 100 1 the the DET fcis-3736 100 2 c_i^pre c_i^pre PROPN fcis-3736 100 3 represent represent VERB fcis-3736 100 4 the the DET fcis-3736 100 5 predicted predict VERB fcis-3736 100 6 number number NOUN fcis-3736 100 7 of of ADP fcis-3736 100 8 people people NOUN fcis-3736 100 9 in in ADP fcis-3736 100 10 the the DET fcis-3736 100 11 i i PROPN fcis-3736 100 12 - - PUNCT fcis-3736 100 13 th th VERB fcis-3736 100 14 image image NOUN fcis-3736 100 15 . . PUNCT fcis-3736 101 1 and and CCONJ fcis-3736 101 2 the the DET fcis-3736 101 3 c_i^gt c_i^gt PROPN fcis-3736 101 4 is be AUX fcis-3736 101 5 the the DET fcis-3736 101 6 ground ground NOUN fcis-3736 101 7 truth truth NOUN fcis-3736 101 8 number number NOUN fcis-3736 101 9 of of ADP fcis-3736 101 10 crowd crowd NOUN fcis-3736 101 11 people people NOUN fcis-3736 101 12 . . PUNCT fcis-3736 102 1 3.2 3.2 NUM fcis-3736 102 2 . . PUNCT fcis-3736 103 1 implementation implementation NOUN fcis-3736 103 2 details detail NOUN fcis-3736 103 3 in in ADP fcis-3736 103 4 the the DET fcis-3736 103 5 training training NOUN fcis-3736 103 6 stage stage NOUN fcis-3736 103 7 of of ADP fcis-3736 103 8 our our PRON fcis-3736 103 9 experiment experiment NOUN fcis-3736 103 10 , , PUNCT fcis-3736 103 11 we we PRON fcis-3736 103 12 set set VERB fcis-3736 103 13 the the DET fcis-3736 103 14 learning learning NOUN fcis-3736 103 15 rate rate NOUN fcis-3736 103 16 at at ADP fcis-3736 103 17 0.00001 0.00001 NUM fcis-3736 103 18 and and CCONJ fcis-3736 103 19 adopt adopt VERB fcis-3736 103 20 adam adam PROPN fcis-3736 103 21 optimizer optimizer NOUN fcis-3736 103 22 . . PUNCT fcis-3736 104 1 in in ADP fcis-3736 104 2 our our PRON fcis-3736 104 3 approach approach NOUN fcis-3736 104 4 , , PUNCT fcis-3736 104 5 the the DET fcis-3736 104 6 size size NOUN fcis-3736 104 7 of of ADP fcis-3736 104 8 input input NOUN fcis-3736 104 9 feature feature NOUN fcis-3736 104 10 is be AUX fcis-3736 104 11 arbitrary arbitrary ADJ fcis-3736 104 12 . . PUNCT fcis-3736 105 1 we we PRON fcis-3736 105 2 set set VERB fcis-3736 105 3 the the DET fcis-3736 105 4 batch batch NOUN fcis-3736 105 5 size size NOUN fcis-3736 105 6 to to ADP fcis-3736 105 7 1 1 NUM fcis-3736 105 8 for for ADP fcis-3736 105 9 the the DET fcis-3736 105 10 datasets dataset NOUN fcis-3736 105 11 that that PRON fcis-3736 105 12 have have VERB fcis-3736 105 13 arbitrary arbitrary ADJ fcis-3736 105 14 size size NOUN fcis-3736 105 15 images image NOUN fcis-3736 105 16 . . PUNCT fcis-3736 106 1 and and CCONJ fcis-3736 106 2 we we PRON fcis-3736 106 3 adopted adopt VERB fcis-3736 106 4 the the DET fcis-3736 106 5 batch batch NOUN fcis-3736 106 6 size size NOUN fcis-3736 106 7 of of ADP fcis-3736 106 8 4 4 NUM fcis-3736 106 9 for for ADP fcis-3736 106 10 the the DET fcis-3736 106 11 shanghai shanghai PROPN fcis-3736 106 12 tech tech PROPN fcis-3736 106 13 _ _ PUNCT fcis-3736 106 14 part part NOUN fcis-3736 106 15 _ _ PUNCT fcis-3736 107 1 b b NOUN fcis-3736 107 2 dataset dataset NOUN fcis-3736 107 3 with with ADP fcis-3736 107 4 fixed fix VERB fcis-3736 107 5 size size NOUN fcis-3736 107 6 images image NOUN fcis-3736 107 7 . . PUNCT fcis-3736 108 1 for for ADP fcis-3736 108 2 each each DET fcis-3736 108 3 dataset dataset NOUN fcis-3736 108 4 , , PUNCT fcis-3736 108 5 we we PRON fcis-3736 108 6 applied apply VERB fcis-3736 108 7 300 300 NUM fcis-3736 108 8 epochs epoch NOUN fcis-3736 108 9 training train VERB fcis-3736 108 10 our our PRON fcis-3736 108 11 model model NOUN fcis-3736 108 12 . . PUNCT fcis-3736 109 1 we we PRON fcis-3736 109 2 use use VERB fcis-3736 109 3 pytorch1.7 pytorch1.7 ADV fcis-3736 109 4 as as ADP fcis-3736 109 5 deep deep ADJ fcis-3736 109 6 learning learning NOUN fcis-3736 109 7 framework framework NOUN fcis-3736 109 8 to to PART fcis-3736 109 9 implement implement VERB fcis-3736 109 10 our our PRON fcis-3736 109 11 method method NOUN fcis-3736 109 12 . . PUNCT fcis-3736 110 1 nvidia nvidia PROPN fcis-3736 110 2 3090 3090 NUM fcis-3736 110 3 gpu gpu NOUN fcis-3736 110 4 is be AUX fcis-3736 110 5 the the DET fcis-3736 110 6 hardware hardware NOUN fcis-3736 110 7 platform platform NOUN fcis-3736 110 8 of of ADP fcis-3736 110 9 our our PRON fcis-3736 110 10 experiment experiment NOUN fcis-3736 110 11 . . PUNCT fcis-3736 111 1 3.3 3.3 NUM fcis-3736 111 2 . . PUNCT fcis-3736 112 1 experimental experimental ADJ fcis-3736 112 2 comparison comparison NOUN fcis-3736 112 3 we we PRON fcis-3736 112 4 test test VERB fcis-3736 112 5 our our PRON fcis-3736 112 6 network network NOUN fcis-3736 112 7 on on ADP fcis-3736 112 8 three three NUM fcis-3736 112 9 publicly publicly ADV fcis-3736 112 10 dataset dataset ADJ fcis-3736 112 11 : : PUNCT fcis-3736 112 12 shanghaies shanghaie NOUN fcis-3736 112 13 , , PUNCT fcis-3736 112 14 ucf ucf PROPN fcis-3736 112 15 - - PUNCT fcis-3736 112 16 cc-50 cc-50 NOUN fcis-3736 112 17 and and CCONJ fcis-3736 112 18 ucf ucf PROPN fcis-3736 112 19 - - PUNCT fcis-3736 112 20 qnrf qnrf NOUN fcis-3736 112 21 . . PUNCT fcis-3736 113 1 the the DET fcis-3736 113 2 summarization summarization NOUN fcis-3736 113 3 of of ADP fcis-3736 113 4 datasets dataset NOUN fcis-3736 113 5 as as SCONJ fcis-3736 113 6 shown show VERB fcis-3736 113 7 in in ADP fcis-3736 113 8 table table NOUN fcis-3736 113 9 1 1 NUM fcis-3736 113 10 . . PUNCT fcis-3736 113 11 and and CCONJ fcis-3736 113 12 the the DET fcis-3736 113 13 comparison comparison NOUN fcis-3736 113 14 with with ADP fcis-3736 113 15 other other ADJ fcis-3736 113 16 experiments experiment NOUN fcis-3736 113 17 are be AUX fcis-3736 113 18 as as SCONJ fcis-3736 113 19 shown show VERB fcis-3736 113 20 in in ADP fcis-3736 113 21 table table NOUN fcis-3736 113 22 2 2 NUM fcis-3736 113 23 . . PUNCT fcis-3736 114 1 and and CCONJ fcis-3736 114 2 in in ADP fcis-3736 114 3 table table NOUN fcis-3736 114 4 2 2 NUM fcis-3736 114 5 , , PUNCT fcis-3736 114 6 denotes denote NOUN fcis-3736 114 7 there there PRON fcis-3736 114 8 is be VERB fcis-3736 114 9 no no DET fcis-3736 114 10 experimental experimental ADJ fcis-3736 114 11 data datum NOUN fcis-3736 114 12 in in ADP fcis-3736 114 13 original original ADJ fcis-3736 114 14 paper paper NOUN fcis-3736 114 15 . . PUNCT fcis-3736 115 1 among among ADP fcis-3736 115 2 these these DET fcis-3736 115 3 datasets dataset NOUN fcis-3736 115 4 , , PUNCT fcis-3736 115 5 we we PRON fcis-3736 115 6 summarized summarize VERB fcis-3736 115 7 that that SCONJ fcis-3736 115 8 shanghaies shanghaie NOUN fcis-3736 115 9 and and CCONJ fcis-3736 115 10 ucf ucf PROPN fcis-3736 115 11 - - PUNCT fcis-3736 115 12 qnrf qnrf PROPN fcis-3736 115 13 have have VERB fcis-3736 115 14 a a DET fcis-3736 115 15 medium medium ADJ fcis-3736 115 16 - - PUNCT fcis-3736 115 17 level level NOUN fcis-3736 115 18 density density NOUN fcis-3736 115 19 crowd crowd NOUN fcis-3736 115 20 distribution distribution NOUN fcis-3736 115 21 . . PUNCT fcis-3736 116 1 ucf ucf PROPN fcis-3736 116 2 - - PUNCT fcis-3736 116 3 cc-50 cc-50 NOUN fcis-3736 116 4 dataset dataset NOUN fcis-3736 116 5 has have VERB fcis-3736 116 6 the the DET fcis-3736 116 7 largest large ADJ fcis-3736 116 8 average average ADJ fcis-3736 116 9 number number NOUN fcis-3736 116 10 of of ADP fcis-3736 116 11 people people NOUN fcis-3736 116 12 in in ADP fcis-3736 116 13 the the DET fcis-3736 116 14 images image NOUN fcis-3736 116 15 . . PUNCT fcis-3736 117 1 shanghaies shanghaie NOUN fcis-3736 117 2 has have VERB fcis-3736 117 3 a a DET fcis-3736 117 4 relatively relatively ADV fcis-3736 117 5 sparse sparse ADJ fcis-3736 117 6 crowd crowd NOUN fcis-3736 117 7 density density NOUN fcis-3736 117 8 distribution distribution NOUN fcis-3736 117 9 . . PUNCT fcis-3736 118 1 table table NOUN fcis-3736 118 2 1 1 NUM fcis-3736 118 3 display display VERB fcis-3736 118 4 the the DET fcis-3736 118 5 comparison comparison NOUN fcis-3736 118 6 among among ADP fcis-3736 118 7 these these DET fcis-3736 118 8 networks network NOUN fcis-3736 118 9 . . PUNCT fcis-3736 119 1 in in ADP fcis-3736 119 2 summary summary NOUN fcis-3736 119 3 , , PUNCT fcis-3736 119 4 our our PRON fcis-3736 119 5 cmf cmf PROPN fcis-3736 119 6 net net PROPN fcis-3736 119 7 has have VERB fcis-3736 119 8 a a DET fcis-3736 119 9 good good ADJ fcis-3736 119 10 performance performance NOUN fcis-3736 119 11 for for ADP fcis-3736 119 12 sparse sparse ADJ fcis-3736 119 13 and and CCONJ fcis-3736 119 14 medium medium ADJ fcis-3736 119 15 - - PUNCT fcis-3736 119 16 level level NOUN fcis-3736 119 17 density density NOUN fcis-3736 119 18 scenes scene NOUN fcis-3736 119 19 . . PUNCT fcis-3736 120 1 3.4 3.4 NUM fcis-3736 120 2 . . PUNCT fcis-3736 120 3 ablation ablation NOUN fcis-3736 120 4 study study NOUN fcis-3736 120 5 in in ADP fcis-3736 120 6 this this DET fcis-3736 120 7 section section NOUN fcis-3736 120 8 , , PUNCT fcis-3736 120 9 we we PRON fcis-3736 120 10 analyzed analyze VERB fcis-3736 120 11 the the DET fcis-3736 120 12 effectiveness effectiveness NOUN fcis-3736 120 13 of of ADP fcis-3736 120 14 each each DET fcis-3736 120 15 module module NOUN fcis-3736 120 16 of of ADP fcis-3736 120 17 our our PRON fcis-3736 120 18 network network NOUN fcis-3736 120 19 . . PUNCT fcis-3736 121 1 we we PRON fcis-3736 121 2 have have VERB fcis-3736 121 3 ablation ablation NOUN fcis-3736 121 4 study study NOUN fcis-3736 121 5 on on ADP fcis-3736 121 6 shanghaies shanghaie NOUN fcis-3736 121 7 dataset dataset VERB fcis-3736 121 8 . . PUNCT fcis-3736 122 1 the the DET fcis-3736 122 2 experimental experimental ADJ fcis-3736 122 3 comparisons comparison NOUN fcis-3736 122 4 are be AUX fcis-3736 122 5 in in ADP fcis-3736 122 6 table table NOUN fcis-3736 122 7 3 3 NUM fcis-3736 122 8 . . PUNCT fcis-3736 122 9 table table NOUN fcis-3736 122 10 1 1 NUM fcis-3736 122 11 . . PUNCT fcis-3736 123 1 the the DET fcis-3736 123 2 datasets dataset NOUN fcis-3736 123 3 summarization summarization NOUN fcis-3736 123 4 datasets dataset NOUN fcis-3736 123 5 resolution resolution NOUN fcis-3736 123 6 train train NOUN fcis-3736 123 7 set set NOUN fcis-3736 123 8 test test NOUN fcis-3736 123 9 set set VERB fcis-3736 123 10 max max PROPN fcis-3736 123 11 min min PROPN fcis-3736 123 12 avg avg PROPN fcis-3736 123 13 sha sha PROPN fcis-3736 123 14 arbitrary arbitrary ADJ fcis-3736 123 15 300 300 NUM fcis-3736 123 16 182 182 NUM fcis-3736 123 17 3,139 3,139 NUM fcis-3736 123 18 33 33 NUM fcis-3736 123 19 501 501 NUM fcis-3736 123 20 shb shb NOUN fcis-3736 123 21 1024 1024 NUM fcis-3736 123 22 * * NOUN fcis-3736 123 23 768 768 NUM fcis-3736 123 24 400 400 NUM fcis-3736 123 25 316 316 NUM fcis-3736 123 26 578 578 NUM fcis-3736 123 27 9 9 NUM fcis-3736 123 28 123 123 NUM fcis-3736 123 29 ucf50 ucf50 NOUN fcis-3736 123 30 arbitrary arbitrary ADJ fcis-3736 123 31 50 50 NUM fcis-3736 123 32 50 50 NUM fcis-3736 123 33 4,543 4,543 NUM fcis-3736 123 34 94 94 NUM fcis-3736 123 35 1,279 1,279 NUM fcis-3736 123 36 qnrf qnrf NOUN fcis-3736 123 37 arbitrary arbitrary ADJ fcis-3736 123 38 1201 1201 NUM fcis-3736 123 39 334 334 NUM fcis-3736 123 40 12,865 12,865 NUM fcis-3736 123 41 49 49 NUM fcis-3736 123 42 815 815 NUM fcis-3736 123 43 table table NOUN fcis-3736 123 44 2 2 NUM fcis-3736 123 45 . . PUNCT fcis-3736 124 1 the the DET fcis-3736 124 2 datasets dataset NOUN fcis-3736 124 3 summarization summarization NOUN fcis-3736 124 4 method method NOUN fcis-3736 124 5 shanghaitech_a shanghaitech_a VERB fcis-3736 124 6 shanghaitech_b shanghaitech_b VERB fcis-3736 124 7 ucf ucf PROPN fcis-3736 124 8 - - PUNCT fcis-3736 124 9 cc-50 cc-50 NOUN fcis-3736 124 10 ucf ucf PROPN fcis-3736 124 11 - - PUNCT fcis-3736 124 12 qnrf qnrf PROPN fcis-3736 124 13 mae mae PROPN fcis-3736 124 14 mse mse PROPN fcis-3736 124 15 mae mae PROPN fcis-3736 124 16 mse mse PROPN fcis-3736 124 17 mae mae PROPN fcis-3736 124 18 mse mse PROPN fcis-3736 124 19 mae mae PROPN fcis-3736 124 20 mse mse PROPN fcis-3736 124 21 mcnn mcnn PROPN fcis-3736 124 22 110.2 110.2 NUM fcis-3736 125 1 173.2 173.2 NUM fcis-3736 125 2 26.4 26.4 NUM fcis-3736 125 3 41.3 41.3 NUM fcis-3736 125 4 377.6 377.6 NUM fcis-3736 125 5 509.1 509.1 NUM fcis-3736 125 6 277 277 NUM fcis-3736 125 7 426 426 NUM fcis-3736 125 8 mscnn mscnn VERB fcis-3736 125 9 83.8 83.8 NUM fcis-3736 125 10 127.4 127.4 NUM fcis-3736 125 11 17.7 17.7 NUM fcis-3736 125 12 30.2 30.2 NUM fcis-3736 125 13 363.7 363.7 NUM fcis-3736 125 14 468.4 468.4 NUM fcis-3736 125 15 swichcnn swichcnn VERB fcis-3736 125 16 90.4 90.4 NUM fcis-3736 125 17 135.0 135.0 NUM fcis-3736 125 18 21.6 21.6 NUM fcis-3736 125 19 33.4 33.4 NUM fcis-3736 125 20 318.1 318.1 NUM fcis-3736 125 21 439.2 439.2 NUM fcis-3736 125 22 228.0 228.0 NUM fcis-3736 125 23 445.0 445.0 NUM fcis-3736 125 24 csrnet csrnet NOUN fcis-3736 125 25 68.2 68.2 NUM fcis-3736 125 26 115.0 115.0 NUM fcis-3736 125 27 10.6 10.6 NUM fcis-3736 125 28 16.0 16.0 NUM fcis-3736 125 29 266.1 266.1 NUM fcis-3736 125 30 397.5 397.5 NUM fcis-3736 125 31 120.3 120.3 NUM fcis-3736 125 32 208.5 208.5 NUM fcis-3736 125 33 lsc lsc PROPN fcis-3736 125 34 - - PUNCT fcis-3736 125 35 cnn cnn PROPN fcis-3736 125 36 66.4 66.4 NUM fcis-3736 125 37 117.0 117.0 NUM fcis-3736 125 38 8.1 8.1 NUM fcis-3736 125 39 12.7 12.7 NUM fcis-3736 125 40 225.6 225.6 NUM fcis-3736 125 41 302.7 302.7 NUM fcis-3736 125 42 120.5 120.5 NUM fcis-3736 125 43 218.2 218.2 NUM fcis-3736 125 44 mdcount mdcount NOUN fcis-3736 125 45 84.2 84.2 NUM fcis-3736 125 46 130.7 130.7 NUM fcis-3736 125 47 11.8 11.8 NUM fcis-3736 125 48 19.15 19.15 NUM fcis-3736 125 49 103.1 103.1 NUM fcis-3736 125 50 158.1 158.1 NUM fcis-3736 125 51 111.3 111.3 NUM fcis-3736 125 52 203 203 NUM fcis-3736 125 53 ligmsa ligmsa NOUN fcis-3736 125 54 76.6 76.6 NUM fcis-3736 125 55 121.4 121.4 NUM fcis-3736 125 56 10.9 10.9 NUM fcis-3736 125 57 17.5 17.5 NUM fcis-3736 125 58 231.5 231.5 NUM fcis-3736 125 59 339.7 339.7 NUM fcis-3736 125 60 ours ours PRON fcis-3736 125 61 65.2 65.2 NUM fcis-3736 125 62 110.3 110.3 NUM fcis-3736 125 63 8.2 8.2 NUM fcis-3736 125 64 11.9 11.9 NUM fcis-3736 125 65 166.3 166.3 NUM fcis-3736 125 66 285.7 285.7 NUM fcis-3736 125 67 110.7 110.7 NUM fcis-3736 125 68 201.3 201.3 NUM fcis-3736 125 69 table table NOUN fcis-3736 125 70 3 3 NUM fcis-3736 125 71 . . PUNCT fcis-3736 126 1 the the DET fcis-3736 126 2 effects effect NOUN fcis-3736 126 3 of of ADP fcis-3736 126 4 each each DET fcis-3736 126 5 modules module NOUN fcis-3736 126 6 shanghaitech_a shanghaitech_a VERB fcis-3736 126 7 mae mae PROPN fcis-3736 126 8 mse mse PROPN fcis-3736 126 9 backbone backbone NOUN fcis-3736 126 10 71.1 71.1 NUM fcis-3736 126 11 121.4 121.4 NUM fcis-3736 126 12 backbone+cmfm(mem backbone+cmfm(mem ADJ fcis-3736 126 13 ) ) PUNCT fcis-3736 126 14 70.3 70.3 NUM fcis-3736 126 15 119.0 119.0 NUM fcis-3736 126 16 backbone+cmfm(mem+cem backbone+cmfm(mem+cem NOUN fcis-3736 126 17 ) ) PUNCT fcis-3736 126 18 69.1 69.1 NUM fcis-3736 126 19 118.7 118.7 NUM fcis-3736 126 20 backbone+cmfm*2 backbone+cmfm*2 NOUN fcis-3736 126 21 68.2 68.2 NUM fcis-3736 126 22 116.1 116.1 NUM fcis-3736 126 23 backbone+cmfm*3 backbone+cmfm*3 NOUN fcis-3736 126 24 66.9 66.9 NUM fcis-3736 126 25 114.0 114.0 NUM fcis-3736 126 26 backbone+cmfm*3+adaptively backbone+cmfm*3+adaptively ADV fcis-3736 126 27 dense dense ADJ fcis-3736 126 28 connection connection NOUN fcis-3736 126 29 65.2 65.2 NUM fcis-3736 126 30 110.3 110.3 NUM fcis-3736 126 31 the the DET fcis-3736 126 32 results result NOUN fcis-3736 126 33 shown show VERB fcis-3736 126 34 the the DET fcis-3736 126 35 modules module NOUN fcis-3736 126 36 we we PRON fcis-3736 126 37 proposed propose VERB fcis-3736 126 38 are be AUX fcis-3736 126 39 effective effective ADJ fcis-3736 126 40 for for ADP fcis-3736 126 41 crowd crowd NOUN fcis-3736 126 42 counting counting NOUN fcis-3736 126 43 task task NOUN fcis-3736 126 44 . . PUNCT fcis-3736 127 1 4 4 X fcis-3736 127 2 . . X fcis-3736 127 3 conclusion conclusion NOUN fcis-3736 127 4 in in ADP fcis-3736 127 5 this this DET fcis-3736 127 6 paper paper NOUN fcis-3736 127 7 , , PUNCT fcis-3736 127 8 we we PRON fcis-3736 127 9 present present VERB fcis-3736 127 10 a a DET fcis-3736 127 11 crowd crowd NOUN fcis-3736 127 12 counting count VERB fcis-3736 127 13 architecture architecture NOUN fcis-3736 127 14 named name VERB fcis-3736 127 15 cmf cmf PROPN fcis-3736 127 16 net net NOUN fcis-3736 127 17 . . PUNCT fcis-3736 128 1 to to PART fcis-3736 128 2 solve solve VERB fcis-3736 128 3 the the DET fcis-3736 128 4 problem problem NOUN fcis-3736 128 5 of of ADP fcis-3736 128 6 perspective perspective ADJ fcis-3736 128 7 effect effect NOUN fcis-3736 128 8 , , PUNCT fcis-3736 128 9 we we PRON fcis-3736 128 10 proposed propose VERB fcis-3736 128 11 multi multi ADJ fcis-3736 128 12 - - ADJ fcis-3736 128 13 scale scale ADJ fcis-3736 128 14 feature feature NOUN fcis-3736 128 15 extraction extraction NOUN fcis-3736 128 16 module module NOUN fcis-3736 128 17 and and CCONJ fcis-3736 128 18 fused fuse VERB fcis-3736 128 19 context context NOUN fcis-3736 128 20 information information NOUN fcis-3736 128 21 . . PUNCT fcis-3736 129 1 we we PRON fcis-3736 129 2 adopted adopt VERB fcis-3736 129 3 adaptively adaptively ADV fcis-3736 129 4 dense dense ADJ fcis-3736 129 5 connect connect NOUN fcis-3736 129 6 to to PART fcis-3736 129 7 further far ADV fcis-3736 129 8 promote promote VERB fcis-3736 129 9 information information NOUN fcis-3736 129 10 transmission transmission NOUN fcis-3736 129 11 . . PUNCT fcis-3736 130 1 the the DET fcis-3736 130 2 results result NOUN fcis-3736 130 3 of of ADP fcis-3736 130 4 extensive extensive ADJ fcis-3736 130 5 experiments experiment NOUN fcis-3736 130 6 shown show VERB fcis-3736 130 7 our our PRON fcis-3736 130 8 cmf cmf PROPN fcis-3736 130 9 net net NOUN fcis-3736 130 10 achieves achieve VERB fcis-3736 130 11 competitive competitive ADJ fcis-3736 130 12 performance performance NOUN fcis-3736 130 13 in in ADP fcis-3736 130 14 accuracy accuracy NOUN fcis-3736 130 15 and and CCONJ fcis-3736 130 16 robustness robustness NOUN fcis-3736 130 17 . . PUNCT fcis-3736 131 1 references reference NOUN fcis-3736 131 2 [ [ X fcis-3736 131 3 1 1 NUM fcis-3736 131 4 ] ] SYM fcis-3736 131 5 dollar dollar NOUN fcis-3736 131 6 p p NOUN fcis-3736 131 7 , , PUNCT fcis-3736 131 8 wojek wojek NOUN fcis-3736 131 9 c c NOUN fcis-3736 131 10 , , PUNCT fcis-3736 131 11 schiele schiele PROPN fcis-3736 131 12 b b PROPN fcis-3736 131 13 and and CCONJ fcis-3736 131 14 perona perona NOUN fcis-3736 131 15 p p PROPN fcis-3736 131 16 2012 2012 NUM fcis-3736 131 17 pedestrian pedestrian NOUN fcis-3736 131 18 detection detection NOUN fcis-3736 131 19 : : PUNCT fcis-3736 131 20 an an DET fcis-3736 131 21 evaluation evaluation NOUN fcis-3736 131 22 of of ADP fcis-3736 131 23 the the DET fcis-3736 131 24 state state NOUN fcis-3736 131 25 of of ADP fcis-3736 131 26 the the DET fcis-3736 131 27 art art NOUN fcis-3736 131 28 ieee ieee NOUN fcis-3736 131 29 transactions transaction NOUN fcis-3736 131 30 on on ADP fcis-3736 131 31 pattern pattern NOUN fcis-3736 131 32 analysis analysis NOUN fcis-3736 131 33 and and CCONJ fcis-3736 131 34 machine machine NOUN fcis-3736 131 35 intelligence intelligence NOUN fcis-3736 131 36 34(4 34(4 NOUN fcis-3736 131 37 ) ) PUNCT fcis-3736 131 38 pp pp ADP fcis-3736 131 39 743–761 743–761 NUM fcis-3736 131 40 . . PUNCT fcis-3736 132 1 [ [ X fcis-3736 132 2 2 2 NUM fcis-3736 132 3 ] ] X fcis-3736 132 4 zhang zhang PROPN fcis-3736 132 5 y y PROPN fcis-3736 132 6 , , PUNCT fcis-3736 132 7 zhou zhou PROPN fcis-3736 132 8 d d PROPN fcis-3736 132 9 , , PUNCT fcis-3736 132 10 chen chen PROPN fcis-3736 132 11 s s PROPN fcis-3736 132 12 , , PUNCT fcis-3736 132 13 gao gao PROPN fcis-3736 132 14 s s PART fcis-3736 132 15 and and CCONJ fcis-3736 132 16 ma ma PROPN fcis-3736 132 17 y y PROPN fcis-3736 132 18 2016 2016 NUM fcis-3736 132 19 single single ADJ fcis-3736 132 20 - - PUNCT fcis-3736 132 21 image image NOUN fcis-3736 132 22 crowd crowd NOUN fcis-3736 132 23 counting counting NOUN fcis-3736 132 24 via via ADP fcis-3736 132 25 multi multi ADJ fcis-3736 132 26 - - ADJ fcis-3736 132 27 column column ADJ fcis-3736 132 28 convolutional convolutional ADJ fcis-3736 132 29 neural neural ADJ fcis-3736 132 30 network network NOUN fcis-3736 132 31 . . PUNCT fcis-3736 133 1 proceedings proceeding NOUN fcis-3736 133 2 of of ADP fcis-3736 133 3 the the DET fcis-3736 133 4 ieee ieee NOUN fcis-3736 133 5 conference conference NOUN fcis-3736 133 6 on on ADP fcis-3736 133 7 computer computer NOUN fcis-3736 133 8 vision vision NOUN fcis-3736 133 9 and and CCONJ fcis-3736 133 10 pattern pattern NOUN fcis-3736 133 11 recognition recognition NOUN fcis-3736 133 12 pp pp ADP fcis-3736 133 13 589–597 589–597 NUM fcis-3736 133 14 . . PUNCT fcis-3736 134 1 [ [ X fcis-3736 134 2 3 3 NUM fcis-3736 134 3 ] ] X fcis-3736 134 4 sindagi sindagi NOUN fcis-3736 134 5 a a PRON fcis-3736 134 6 and and CCONJ fcis-3736 134 7 patel patel PROPN fcis-3736 134 8 m m PROPN fcis-3736 134 9 2017 2017 NUM fcis-3736 134 10 generating generate VERB fcis-3736 134 11 high high ADJ fcis-3736 134 12 - - PUNCT fcis-3736 134 13 quality quality NOUN fcis-3736 134 14 crowd crowd NOUN fcis-3736 134 15 density density NOUN fcis-3736 134 16 maps map NOUN fcis-3736 134 17 using use VERB fcis-3736 134 18 contextual contextual ADJ fcis-3736 134 19 pyramid pyramid NOUN fcis-3736 134 20 cnns cnns PROPN fcis-3736 134 21 ieee ieee PROPN fcis-3736 134 22 international international PROPN fcis-3736 134 23 conference conference NOUN fcis-3736 134 24 on on ADP fcis-3736 134 25 computer computer NOUN fcis-3736 134 26 vision vision NOUN fcis-3736 134 27 ( ( PUNCT fcis-3736 134 28 iccv iccv PROPN fcis-3736 134 29 ) ) PUNCT fcis-3736 134 30 pp pp ADP fcis-3736 134 31 1879–1888 1879–1888 NUM fcis-3736 134 32 . . PUNCT fcis-3736 135 1 [ [ X fcis-3736 135 2 4 4 X fcis-3736 135 3 ] ] X fcis-3736 135 4 [ [ X fcis-3736 135 5 4 4 NUM fcis-3736 135 6 ] ] SYM fcis-3736 135 7 li li PROPN fcis-3736 135 8 y y PROPN fcis-3736 135 9 , , PUNCT fcis-3736 135 10 zhang zhang PROPN fcis-3736 135 11 x x PROPN fcis-3736 135 12 and and CCONJ fcis-3736 135 13 chen chen PROPN fcis-3736 135 14 d d PROPN fcis-3736 135 15 2018 2018 NUM fcis-3736 135 16 csrnet csrnet NOUN fcis-3736 135 17 : : PUNCT fcis-3736 135 18 dilated dilate VERB fcis-3736 135 19 convolutional convolutional ADJ fcis-3736 135 20 neural neural ADJ fcis-3736 135 21 networks network NOUN fcis-3736 135 22 for for ADP fcis-3736 135 23 understanding understand VERB fcis-3736 135 24 the the DET fcis-3736 135 25 highly highly ADV fcis-3736 135 26 congested congested ADJ fcis-3736 135 27 scenes scene NOUN fcis-3736 135 28 proceedings proceeding NOUN fcis-3736 135 29 of of ADP fcis-3736 135 30 the the DET fcis-3736 135 31 ieee ieee NOUN fcis-3736 135 32 conference conference NOUN fcis-3736 135 33 on on ADP fcis-3736 135 34 computer computer NOUN fcis-3736 135 35 vision vision NOUN fcis-3736 135 36 and and CCONJ fcis-3736 135 37 pattern pattern NOUN fcis-3736 135 38 recognition recognition NOUN fcis-3736 135 39 pp pp ADP fcis-3736 135 40 1091 1091 NUM fcis-3736 135 41 - - SYM fcis-3736 135 42 1100 1100 NUM fcis-3736 135 43 . . PUNCT fcis-3736 136 1 [ [ X fcis-3736 136 2 5 5 NUM fcis-3736 136 3 ] ] X fcis-3736 136 4 davide davide PROPN fcis-3736 136 5 m m PROPN fcis-3736 136 6 , , PUNCT fcis-3736 136 7 bing bing VERB fcis-3736 136 8 s s PRON fcis-3736 136 9 , , PUNCT fcis-3736 136 10 rahul rahul PROPN fcis-3736 136 11 r r NOUN fcis-3736 136 12 v v PROPN fcis-3736 136 13 , , PUNCT fcis-3736 136 14 joseph joseph PROPN fcis-3736 136 15 t t PROPN fcis-3736 136 16 2021 2021 NUM fcis-3736 136 17 understanding understand VERB fcis-3736 136 18 the the DET fcis-3736 136 19 impact impact NOUN fcis-3736 136 20 of of ADP fcis-3736 136 21 mistakes mistake NOUN fcis-3736 136 22 on on ADP fcis-3736 136 23 background background NOUN fcis-3736 136 24 regions region NOUN fcis-3736 136 25 in in ADP fcis-3736 136 26 crowd crowd NOUN fcis-3736 136 27 counting counting NOUN fcis-3736 136 28 proceedings proceeding NOUN fcis-3736 136 29 of of ADP fcis-3736 136 30 the the DET fcis-3736 136 31 ieee ieee NOUN fcis-3736 136 32 / / SYM fcis-3736 136 33 cvf cvf NOUN fcis-3736 136 34 winter winter NOUN fcis-3736 136 35 conference conference NOUN fcis-3736 136 36 on on ADP fcis-3736 136 37 applications application NOUN fcis-3736 136 38 of of ADP fcis-3736 136 39 computer computer NOUN fcis-3736 136 40 vision vision NOUN fcis-3736 136 41 ( ( PUNCT fcis-3736 136 42 wacv wacv NOUN fcis-3736 136 43 ) ) PUNCT fcis-3736 136 44 , , PUNCT fcis-3736 136 45 2021 2021 NUM fcis-3736 136 46 , , PUNCT fcis-3736 136 47 pp pp ADJ fcis-3736 136 48 . . PUNCT fcis-3736 136 49 16501659 16501659 NUM fcis-3736 136 50 . . PUNCT fcis-3736 137 1 [ [ X fcis-3736 137 2 6 6 NUM fcis-3736 137 3 ] ] X fcis-3736 137 4 karen karen PROPN fcis-3736 137 5 s s PROPN fcis-3736 137 6 and and CCONJ fcis-3736 137 7 andrew andrew PROPN fcis-3736 137 8 z z PROPN fcis-3736 137 9 2014 2014 NUM fcis-3736 137 10 very very ADV fcis-3736 137 11 deep deep ADJ fcis-3736 137 12 convolutional convolutional ADJ fcis-3736 137 13 networks network NOUN fcis-3736 137 14 for for ADP fcis-3736 137 15 large large ADJ fcis-3736 137 16 - - PUNCT fcis-3736 137 17 scale scale NOUN fcis-3736 137 18 image image NOUN fcis-3736 137 19 recognition recognition NOUN fcis-3736 137 20 computer computer NOUN fcis-3736 137 21 vision vision NOUN fcis-3736 137 22 and and CCONJ fcis-3736 137 23 pattern pattern NOUN fcis-3736 137 24 recognition recognition NOUN fcis-3736 137 25 ( ( PUNCT fcis-3736 137 26 cs.cv cs.cv NOUN fcis-3736 137 27 ) ) PUNCT fcis-3736 137 28 arxiv:1409.1556 arxiv:1409.1556 NOUN fcis-3736 137 29 . . PUNCT fcis-3736 138 1 [ [ X fcis-3736 138 2 7 7 NUM fcis-3736 138 3 ] ] X fcis-3736 138 4 weizhe weizhe ADJ fcis-3736 138 5 l l NOUN fcis-3736 138 6 , , PUNCT fcis-3736 138 7 mathieu mathieu PROPN fcis-3736 138 8 s s PART fcis-3736 138 9 and and CCONJ fcis-3736 138 10 pascal pascal PROPN fcis-3736 138 11 f f PROPN fcis-3736 138 12 2019 2019 NUM fcis-3736 138 13 context context NOUN fcis-3736 138 14 - - PUNCT fcis-3736 138 15 aware aware ADJ fcis-3736 138 16 crowd crowd NOUN fcis-3736 138 17 counting counting NOUN fcis-3736 138 18 proceedings proceeding NOUN fcis-3736 138 19 of of ADP fcis-3736 138 20 the the DET fcis-3736 138 21 ieee ieee NOUN fcis-3736 138 22 / / SYM fcis-3736 138 23 cvf cvf NOUN fcis-3736 138 24 conference conference NOUN fcis-3736 138 25 on on ADP fcis-3736 138 26 computer computer NOUN fcis-3736 138 27 vision vision NOUN fcis-3736 138 28 and and CCONJ fcis-3736 138 29 pattern pattern NOUN fcis-3736 138 30 recognition recognition NOUN fcis-3736 138 31 ( ( PUNCT fcis-3736 138 32 cvpr cvpr NOUN fcis-3736 138 33 ) ) PUNCT fcis-3736 138 34 pp pp ADP fcis-3736 138 35 50995108 50995108 NUM fcis-3736 138 36 . . PUNCT fcis-3736 139 1 [ [ X fcis-3736 139 2 8 8 NUM fcis-3736 139 3 ] ] X fcis-3736 139 4 liang liang PROPN fcis-3736 139 5 - - PUNCT fcis-3736 139 6 chieh chieh PROPN fcis-3736 139 7 c c PROPN fcis-3736 139 8 , , PUNCT fcis-3736 139 9 george george PROPN fcis-3736 139 10 p p PROPN fcis-3736 139 11 , , PUNCT fcis-3736 139 12 iasonas iasonas PROPN fcis-3736 139 13 k k PROPN fcis-3736 139 14 , , PUNCT fcis-3736 139 15 kevin kevin PROPN fcis-3736 139 16 m m PROPN fcis-3736 139 17 and and CCONJ fcis-3736 139 18 alan alan PROPN fcis-3736 139 19 l l PROPN fcis-3736 139 20 y y PROPN fcis-3736 139 21 2017 2017 NUM fcis-3736 139 22 deeplab deeplab NOUN fcis-3736 139 23 : : PUNCT fcis-3736 139 24 semantic semantic ADJ fcis-3736 139 25 image image NOUN fcis-3736 139 26 segmentation segmentation NOUN fcis-3736 139 27 with with ADP fcis-3736 139 28 deep deep ADJ fcis-3736 139 29 convolutional convolutional ADJ fcis-3736 139 30 nets net NOUN fcis-3736 139 31 , , PUNCT fcis-3736 139 32 atrous atrous ADJ fcis-3736 139 33 convolution convolution NOUN fcis-3736 139 34 , , PUNCT fcis-3736 139 35 and and CCONJ fcis-3736 139 36 fully fully ADV fcis-3736 139 37 connected connect VERB fcis-3736 139 38 crfs crfs PROPN fcis-3736 139 39 ieee ieee NOUN fcis-3736 139 40 transactions transaction NOUN fcis-3736 139 41 on on ADP fcis-3736 139 42 pattern pattern NOUN fcis-3736 139 43 analysis analysis NOUN fcis-3736 139 44 and and CCONJ fcis-3736 139 45 machine machine NOUN fcis-3736 139 46 intelligence intelligence NOUN fcis-3736 139 47 , , PUNCT fcis-3736 139 48 2017 2017 NUM fcis-3736 139 49 , , PUNCT fcis-3736 139 50 40(4 40(4 NUM fcis-3736 139 51 ): ): PUNCT fcis-3736 139 52 834 834 NUM fcis-3736 139 53 - - SYM fcis-3736 139 54 848 848 NUM fcis-3736 139 55 . . PUNCT fcis-3736 139 56 15 15 NUM fcis-3736 140 1 [ [ X fcis-3736 140 2 9 9 NUM fcis-3736 140 3 ] ] PUNCT fcis-3736 140 4 sanghyun sanghyun NOUN fcis-3736 140 5 w w PROPN fcis-3736 140 6 , , PUNCT fcis-3736 140 7 jongchan jongchan ADP fcis-3736 140 8 p p X fcis-3736 140 9 , , PUNCT fcis-3736 140 10 joon joon NOUN fcis-3736 140 11 - - PUNCT fcis-3736 140 12 young young ADJ fcis-3736 140 13 l l NOUN fcis-3736 140 14 and and CCONJ fcis-3736 140 15 in in ADP fcis-3736 140 16 s s PROPN fcis-3736 140 17 k k PROPN fcis-3736 140 18 2018 2018 NUM fcis-3736 140 19 cbam cbam NOUN fcis-3736 140 20 : : PUNCT fcis-3736 140 21 convolutional convolutional ADJ fcis-3736 140 22 block block NOUN fcis-3736 140 23 attention attention NOUN fcis-3736 140 24 module module NOUN fcis-3736 140 25 proceedings proceeding NOUN fcis-3736 140 26 of of ADP fcis-3736 140 27 the the DET fcis-3736 140 28 european european PROPN fcis-3736 140 29 conference conference PROPN fcis-3736 140 30 on on ADP fcis-3736 140 31 computer computer NOUN fcis-3736 140 32 vision vision NOUN fcis-3736 140 33 ( ( PUNCT fcis-3736 140 34 eccv eccv ADV fcis-3736 140 35 ) ) PUNCT fcis-3736 140 36 , , PUNCT fcis-3736 140 37 pp pp ADP fcis-3736 140 38 . . PUNCT fcis-3736 141 1 3 3 NUM fcis-3736 141 2 - - SYM fcis-3736 141 3 19 19 NUM fcis-3736 141 4 . . PUNCT fcis-3736 142 1 [ [ X fcis-3736 142 2 10 10 NUM fcis-3736 142 3 ] ] X fcis-3736 142 4 gao gao PROPN fcis-3736 142 5 h h PROPN fcis-3736 142 6 , , PUNCT fcis-3736 142 7 zhuang zhuang PROPN fcis-3736 142 8 l l PROPN fcis-3736 142 9 , , PUNCT fcis-3736 142 10 laurens laurens PROPN fcis-3736 142 11 van van PROPN fcis-3736 142 12 der der PROPN fcis-3736 142 13 m m PROPN fcis-3736 142 14 and and CCONJ fcis-3736 142 15 kilian kilian PROPN fcis-3736 142 16 q. q. PROPN fcis-3736 142 17 w w PROPN fcis-3736 142 18 2017 2017 NUM fcis-3736 142 19 densely densely ADV fcis-3736 142 20 connected connect VERB fcis-3736 142 21 convolutional convolutional ADJ fcis-3736 142 22 networks network NOUN fcis-3736 142 23 proceedings proceeding NOUN fcis-3736 142 24 of of ADP fcis-3736 142 25 the the DET fcis-3736 142 26 ieee ieee NOUN fcis-3736 142 27 conference conference NOUN fcis-3736 142 28 on on ADP fcis-3736 142 29 computer computer NOUN fcis-3736 142 30 vision vision NOUN fcis-3736 142 31 and and CCONJ fcis-3736 142 32 pattern pattern NOUN fcis-3736 142 33 recognition recognition NOUN fcis-3736 142 34 ( ( PUNCT fcis-3736 142 35 cvpr cvpr NOUN fcis-3736 142 36 ) ) PUNCT fcis-3736 142 37 pp pp ADP fcis-3736 142 38 . . PUNCT fcis-3736 142 39 4700 4700 NUM fcis-3736 142 40 - - SYM fcis-3736 142 41 4708 4708 NUM fcis-3736 142 42 . . PUNCT