id sid tid token lemma pos fcis-6350 1 1 frontiers frontier NOUN fcis-6350 1 2 in in ADP fcis-6350 1 3 computing computing NOUN fcis-6350 1 4 and and CCONJ fcis-6350 1 5 intelligent intelligent ADJ fcis-6350 1 6 systems system NOUN fcis-6350 1 7 issn issn VERB fcis-6350 1 8 : : PUNCT fcis-6350 1 9 2832 2832 NUM fcis-6350 1 10 - - SYM fcis-6350 1 11 6024 6024 NUM fcis-6350 1 12 | | NOUN fcis-6350 1 13 vol vol NOUN fcis-6350 1 14 . . PROPN fcis-6350 2 1 3 3 NUM fcis-6350 2 2 , , PUNCT fcis-6350 2 3 no no INTJ fcis-6350 2 4 . . NOUN fcis-6350 2 5 1 1 NUM fcis-6350 2 6 , , PUNCT fcis-6350 2 7 2023 2023 NUM fcis-6350 2 8 134 134 NUM fcis-6350 2 9 road road NOUN fcis-6350 2 10 traffic traffic NOUN fcis-6350 2 11 small small ADJ fcis-6350 2 12 target target NOUN fcis-6350 2 13 detection detection NOUN fcis-6350 2 14 algorithm algorithm NOUN fcis-6350 2 15 based base VERB fcis-6350 2 16 on on ADP fcis-6350 2 17 lightweight lightweight ADJ fcis-6350 2 18 yolo yolo ADJ fcis-6350 2 19 v5 v5 PROPN fcis-6350 2 20 chanyi chanyi PROPN fcis-6350 2 21 liu liu PROPN fcis-6350 2 22 1 1 NUM fcis-6350 2 23 , , PUNCT fcis-6350 2 24 2 2 NUM fcis-6350 2 25 * * NUM fcis-6350 2 26 , , PUNCT fcis-6350 2 27 dan dan PROPN fcis-6350 2 28 huang huang PROPN fcis-6350 2 29 1 1 NUM fcis-6350 2 30 , , PUNCT fcis-6350 2 31 2 2 NUM fcis-6350 2 32 , , PUNCT fcis-6350 2 33 tao tao PROPN fcis-6350 2 34 wang wang PROPN fcis-6350 2 35 1 1 NUM fcis-6350 2 36 , , PUNCT fcis-6350 2 37 2 2 NUM fcis-6350 2 38 , , PUNCT fcis-6350 2 39 tao tao PROPN fcis-6350 2 40 zhu zhu PROPN fcis-6350 2 41 1 1 NUM fcis-6350 2 42 , , PUNCT fcis-6350 2 43 2 2 NUM fcis-6350 2 44 1 1 NUM fcis-6350 2 45 school school NOUN fcis-6350 2 46 of of ADP fcis-6350 2 47 automation automation NOUN fcis-6350 2 48 and and CCONJ fcis-6350 2 49 information information NOUN fcis-6350 2 50 engineering engineering NOUN fcis-6350 2 51 , , PUNCT fcis-6350 2 52 sichuan sichuan PROPN fcis-6350 2 53 university university PROPN fcis-6350 2 54 of of ADP fcis-6350 2 55 science science PROPN fcis-6350 2 56 & & CCONJ fcis-6350 2 57 engineering engineering PROPN fcis-6350 2 58 , , PUNCT fcis-6350 2 59 yibin yibin PROPN fcis-6350 2 60 , , PUNCT fcis-6350 2 61 644002 644002 NUM fcis-6350 2 62 , , PUNCT fcis-6350 2 63 china china PROPN fcis-6350 2 64 2artificial 2artificial PROPN fcis-6350 2 65 intelligence intelligence PROPN fcis-6350 2 66 key key ADJ fcis-6350 2 67 laboratory laboratory NOUN fcis-6350 2 68 of of ADP fcis-6350 2 69 sichuan sichuan PROPN fcis-6350 2 70 province province PROPN fcis-6350 2 71 , , PUNCT fcis-6350 2 72 sichuan sichuan PROPN fcis-6350 2 73 university university PROPN fcis-6350 2 74 of of ADP fcis-6350 2 75 science science PROPN fcis-6350 2 76 & & CCONJ fcis-6350 2 77 engineering engineering PROPN fcis-6350 2 78 , , PUNCT fcis-6350 2 79 yibin yibin PROPN fcis-6350 2 80 , , PUNCT fcis-6350 2 81 644002 644002 NUM fcis-6350 2 82 , , PUNCT fcis-6350 2 83 china china PROPN fcis-6350 2 84 ; ; PUNCT fcis-6350 2 85 * * PUNCT fcis-6350 2 86 corresponding correspond VERB fcis-6350 2 87 author author NOUN fcis-6350 2 88 : : PUNCT fcis-6350 2 89 chanyi chanyi PROPN fcis-6350 2 90 liu liu PROPN fcis-6350 2 91 ( ( PUNCT fcis-6350 2 92 email email NOUN fcis-6350 2 93 : : PUNCT fcis-6350 2 94 2720806204@qq.com 2720806204@qq.com NUM fcis-6350 2 95 ) ) PUNCT fcis-6350 2 96 abstract abstract NOUN fcis-6350 2 97 : : PUNCT fcis-6350 2 98 for for ADP fcis-6350 2 99 the the DET fcis-6350 2 100 current current ADJ fcis-6350 2 101 problem problem NOUN fcis-6350 2 102 of of ADP fcis-6350 2 103 large large ADJ fcis-6350 2 104 number number NOUN fcis-6350 2 105 of of ADP fcis-6350 2 106 parameters parameter NOUN fcis-6350 2 107 and and CCONJ fcis-6350 2 108 large large ADJ fcis-6350 2 109 computation computation NOUN fcis-6350 2 110 of of ADP fcis-6350 2 111 the the DET fcis-6350 2 112 target target NOUN fcis-6350 2 113 detection detection NOUN fcis-6350 2 114 network network NOUN fcis-6350 2 115 , , PUNCT fcis-6350 2 116 the the DET fcis-6350 2 117 lightness lightness ADJ fcis-6350 2 118 improvement improvement NOUN fcis-6350 2 119 is be AUX fcis-6350 2 120 carried carry VERB fcis-6350 2 121 out out ADP fcis-6350 2 122 on on ADP fcis-6350 2 123 the the DET fcis-6350 2 124 basis basis NOUN fcis-6350 2 125 of of ADP fcis-6350 2 126 yolov5s yolov5s PROPN fcis-6350 2 127 , , PUNCT fcis-6350 2 128 and and CCONJ fcis-6350 2 129 the the DET fcis-6350 2 130 dynamic dynamic ADJ fcis-6350 2 131 convolution convolution NOUN fcis-6350 2 132 of of ADP fcis-6350 2 133 od od PROPN fcis-6350 2 134 is be AUX fcis-6350 2 135 introduced introduce VERB fcis-6350 2 136 to to PART fcis-6350 2 137 compensate compensate VERB fcis-6350 2 138 for for ADP fcis-6350 2 139 the the DET fcis-6350 2 140 accuracy accuracy NOUN fcis-6350 2 141 degradation degradation NOUN fcis-6350 2 142 caused cause VERB fcis-6350 2 143 by by ADP fcis-6350 2 144 the the DET fcis-6350 2 145 lightness lightness NOUN fcis-6350 2 146 , , PUNCT fcis-6350 2 147 and and CCONJ fcis-6350 2 148 the the DET fcis-6350 2 149 map0.5 map0.5 PROPN fcis-6350 2 150 of of ADP fcis-6350 2 151 the the DET fcis-6350 2 152 improved improved ADJ fcis-6350 2 153 yolov5sefficientnetv2 yolov5sefficientnetv2 PROPN fcis-6350 2 154 - - PUNCT fcis-6350 2 155 od od PROPN fcis-6350 2 156 network network NOUN fcis-6350 2 157 reaches reach VERB fcis-6350 2 158 82.8 82.8 NUM fcis-6350 2 159 through through ADP fcis-6350 2 160 comparison comparison NOUN fcis-6350 2 161 experiments experiment NOUN fcis-6350 2 162 , , PUNCT fcis-6350 2 163 and and CCONJ fcis-6350 2 164 the the DET fcis-6350 2 165 fps fps NOUN fcis-6350 2 166 is be AUX fcis-6350 2 167 59 59 NUM fcis-6350 2 168 frames frame NOUN fcis-6350 2 169 per per ADP fcis-6350 2 170 second second NOUN fcis-6350 2 171 . . PUNCT fcis-6350 3 1 the the DET fcis-6350 3 2 experiment experiment NOUN fcis-6350 3 3 proves prove VERB fcis-6350 3 4 that that SCONJ fcis-6350 3 5 the the DET fcis-6350 3 6 network network NOUN fcis-6350 3 7 maintains maintain VERB fcis-6350 3 8 the the DET fcis-6350 3 9 detection detection NOUN fcis-6350 3 10 accuracy accuracy NOUN fcis-6350 3 11 and and CCONJ fcis-6350 3 12 detection detection NOUN fcis-6350 3 13 speed speed NOUN fcis-6350 3 14 of of ADP fcis-6350 3 15 the the DET fcis-6350 3 16 network network NOUN fcis-6350 3 17 for for ADP fcis-6350 3 18 small small ADJ fcis-6350 3 19 targets target NOUN fcis-6350 3 20 on on ADP fcis-6350 3 21 the the DET fcis-6350 3 22 basis basis NOUN fcis-6350 3 23 of of ADP fcis-6350 3 24 light light ADJ fcis-6350 3 25 weighting weighting NOUN fcis-6350 3 26 . . PUNCT fcis-6350 4 1 it it PRON fcis-6350 4 2 is be AUX fcis-6350 4 3 suitable suitable ADJ fcis-6350 4 4 for for ADP fcis-6350 4 5 deployment deployment NOUN fcis-6350 4 6 on on ADP fcis-6350 4 7 vehicle vehicle NOUN fcis-6350 4 8 - - PUNCT fcis-6350 4 9 mounted mount VERB fcis-6350 4 10 devices device NOUN fcis-6350 4 11 . . PUNCT fcis-6350 5 1 keywords keyword NOUN fcis-6350 5 2 : : PUNCT fcis-6350 5 3 deep deep ADJ fcis-6350 5 4 learning learning NOUN fcis-6350 5 5 ; ; PUNCT fcis-6350 5 6 lightweight lightweight ADJ fcis-6350 5 7 ; ; PUNCT fcis-6350 5 8 small small ADJ fcis-6350 5 9 target target NOUN fcis-6350 5 10 detection detection NOUN fcis-6350 5 11 . . PUNCT fcis-6350 6 1 1 1 X fcis-6350 6 2 . . X fcis-6350 6 3 introduction introduction NOUN fcis-6350 6 4 with with ADP fcis-6350 6 5 the the DET fcis-6350 6 6 rise rise NOUN fcis-6350 6 7 in in ADP fcis-6350 6 8 per per ADP fcis-6350 6 9 capita capita X fcis-6350 6 10 household household NOUN fcis-6350 6 11 income income NOUN fcis-6350 6 12 in in ADP fcis-6350 6 13 china china PROPN fcis-6350 6 14 , , PUNCT fcis-6350 6 15 motor motor NOUN fcis-6350 6 16 vehicle vehicle NOUN fcis-6350 6 17 ownership ownership NOUN fcis-6350 6 18 has have AUX fcis-6350 6 19 increased increase VERB fcis-6350 6 20 dramatically dramatically ADV fcis-6350 6 21 , , PUNCT fcis-6350 6 22 and and CCONJ fcis-6350 6 23 the the DET fcis-6350 6 24 increase increase NOUN fcis-6350 6 25 in in ADP fcis-6350 6 26 vehicle vehicle NOUN fcis-6350 6 27 ownership ownership NOUN fcis-6350 6 28 has have AUX fcis-6350 6 29 brought bring VERB fcis-6350 6 30 great great ADJ fcis-6350 6 31 pressure pressure NOUN fcis-6350 6 32 on on ADP fcis-6350 6 33 the the DET fcis-6350 6 34 roads road NOUN fcis-6350 6 35 and and CCONJ fcis-6350 6 36 the the DET fcis-6350 6 37 ensuing ensue VERB fcis-6350 6 38 traffic traffic NOUN fcis-6350 6 39 accidents accident NOUN fcis-6350 6 40 . . PUNCT fcis-6350 7 1 in in ADP fcis-6350 7 2 addition addition NOUN fcis-6350 7 3 to to ADP fcis-6350 7 4 the the DET fcis-6350 7 5 weather weather NOUN fcis-6350 7 6 and and CCONJ fcis-6350 7 7 the the DET fcis-6350 7 8 vehicle vehicle NOUN fcis-6350 7 9 itself itself PRON fcis-6350 7 10 mechanical mechanical ADJ fcis-6350 7 11 failure failure NOUN fcis-6350 7 12 and and CCONJ fcis-6350 7 13 other other ADJ fcis-6350 7 14 objective objective ADJ fcis-6350 7 15 factors factor NOUN fcis-6350 7 16 , , PUNCT fcis-6350 7 17 traditional traditional ADJ fcis-6350 7 18 car car NOUN fcis-6350 7 19 driving driving NOUN fcis-6350 7 20 requires require VERB fcis-6350 7 21 the the DET fcis-6350 7 22 driver driver NOUN fcis-6350 7 23 to to PART fcis-6350 7 24 always always ADV fcis-6350 7 25 observe observe VERB fcis-6350 7 26 the the DET fcis-6350 7 27 surrounding surround VERB fcis-6350 7 28 road road NOUN fcis-6350 7 29 information information NOUN fcis-6350 7 30 , , PUNCT fcis-6350 7 31 if if SCONJ fcis-6350 7 32 fatigue fatigue NOUN fcis-6350 7 33 , , PUNCT fcis-6350 7 34 distracted distracted AUX fcis-6350 7 35 easily easily ADV fcis-6350 7 36 caused cause VERB fcis-6350 7 37 by by ADP fcis-6350 7 38 traffic traffic NOUN fcis-6350 7 39 accidents accident NOUN fcis-6350 7 40 . . PUNCT fcis-6350 8 1 with with ADP fcis-6350 8 2 the the DET fcis-6350 8 3 development development NOUN fcis-6350 8 4 of of ADP fcis-6350 8 5 artificial artificial ADJ fcis-6350 8 6 intelligence intelligence NOUN fcis-6350 8 7 , , PUNCT fcis-6350 8 8 the the DET fcis-6350 8 9 technology technology NOUN fcis-6350 8 10 of of ADP fcis-6350 8 11 automatic automatic ADJ fcis-6350 8 12 driving driving NOUN fcis-6350 8 13 is be AUX fcis-6350 8 14 becoming become VERB fcis-6350 8 15 more more ADV fcis-6350 8 16 and and CCONJ fcis-6350 8 17 more more ADV fcis-6350 8 18 mature mature ADJ fcis-6350 8 19 . . PUNCT fcis-6350 9 1 there there PRON fcis-6350 9 2 is be VERB fcis-6350 9 3 a a DET fcis-6350 9 4 big big ADJ fcis-6350 9 5 difference difference NOUN fcis-6350 9 6 between between ADP fcis-6350 9 7 traditional traditional ADJ fcis-6350 9 8 manual manual ADJ fcis-6350 9 9 driving drive VERB fcis-6350 9 10 vehicles vehicle NOUN fcis-6350 9 11 and and CCONJ fcis-6350 9 12 self self NOUN fcis-6350 9 13 - - PUNCT fcis-6350 9 14 driving drive VERB fcis-6350 9 15 vehicles vehicle NOUN fcis-6350 9 16 , , PUNCT fcis-6350 9 17 and and CCONJ fcis-6350 9 18 automatic automatic ADJ fcis-6350 9 19 driving driving NOUN fcis-6350 9 20 can can AUX fcis-6350 9 21 be be AUX fcis-6350 9 22 achieved achieve VERB fcis-6350 9 23 without without ADP fcis-6350 9 24 human human ADJ fcis-6350 9 25 intervention intervention NOUN fcis-6350 9 26 and and CCONJ fcis-6350 9 27 precise precise ADJ fcis-6350 9 28 automatic automatic ADJ fcis-6350 9 29 control control NOUN fcis-6350 9 30 , , PUNCT fcis-6350 9 31 and and CCONJ fcis-6350 9 32 when when SCONJ fcis-6350 9 33 there there PRON fcis-6350 9 34 is be VERB fcis-6350 9 35 an an DET fcis-6350 9 36 unexpected unexpected ADJ fcis-6350 9 37 situation situation NOUN fcis-6350 9 38 , , PUNCT fcis-6350 9 39 the the DET fcis-6350 9 40 automatic automatic ADJ fcis-6350 9 41 driving driving NOUN fcis-6350 9 42 system system NOUN fcis-6350 9 43 will will AUX fcis-6350 9 44 intervene intervene VERB fcis-6350 9 45 in in ADP fcis-6350 9 46 advance advance NOUN fcis-6350 9 47 as as ADV fcis-6350 9 48 well well ADV fcis-6350 9 49 as as ADP fcis-6350 9 50 emergency emergency NOUN fcis-6350 9 51 braking braking NOUN fcis-6350 9 52 when when SCONJ fcis-6350 9 53 necessary necessary ADJ fcis-6350 9 54 . . PUNCT fcis-6350 10 1 as as ADP fcis-6350 10 2 a a DET fcis-6350 10 3 basic basic ADJ fcis-6350 10 4 component component NOUN fcis-6350 10 5 of of ADP fcis-6350 10 6 an an DET fcis-6350 10 7 autonomous autonomous ADJ fcis-6350 10 8 driving driving NOUN fcis-6350 10 9 system system NOUN fcis-6350 10 10 , , PUNCT fcis-6350 10 11 the the DET fcis-6350 10 12 detection detection NOUN fcis-6350 10 13 capability capability NOUN fcis-6350 10 14 of of ADP fcis-6350 10 15 the the DET fcis-6350 10 16 target target NOUN fcis-6350 10 17 detection detection NOUN fcis-6350 10 18 is be AUX fcis-6350 10 19 one one NUM fcis-6350 10 20 of of ADP fcis-6350 10 21 the the DET fcis-6350 10 22 reference reference NOUN fcis-6350 10 23 indicators indicator NOUN fcis-6350 10 24 for for ADP fcis-6350 10 25 judging judge VERB fcis-6350 10 26 the the DET fcis-6350 10 27 level level NOUN fcis-6350 10 28 of of ADP fcis-6350 10 29 autonomous autonomous ADJ fcis-6350 10 30 driving driving NOUN fcis-6350 10 31 . . PUNCT fcis-6350 11 1 common common ADJ fcis-6350 11 2 detection detection NOUN fcis-6350 11 3 networks network NOUN fcis-6350 11 4 are be AUX fcis-6350 11 5 divided divide VERB fcis-6350 11 6 into into ADP fcis-6350 11 7 twostage twostage NOUN fcis-6350 11 8 detection detection NOUN fcis-6350 11 9 networks network NOUN fcis-6350 11 10 such such ADJ fcis-6350 11 11 as as ADP fcis-6350 11 12 rcnn rcnn NOUN fcis-6350 11 13 , , PUNCT fcis-6350 11 14 fast fast ADJ fcis-6350 11 15 - - PUNCT fcis-6350 11 16 rcnn rcnn NOUN fcis-6350 11 17 , , PUNCT fcis-6350 11 18 fasterrcnn fasterrcnn PROPN fcis-6350 11 19 , , PUNCT fcis-6350 11 20 single single ADJ fcis-6350 11 21 - - PUNCT fcis-6350 11 22 stage stage NOUN fcis-6350 11 23 detection detection NOUN fcis-6350 11 24 networks network NOUN fcis-6350 11 25 ssd ssd NOUN fcis-6350 11 26 , , PUNCT fcis-6350 11 27 and and CCONJ fcis-6350 11 28 yolo[5 yolo[5 PROPN fcis-6350 11 29 - - SYM fcis-6350 11 30 9 9 NUM fcis-6350 11 31 ] ] PUNCT fcis-6350 11 32 series series NOUN fcis-6350 11 33 . . PUNCT fcis-6350 12 1 the the DET fcis-6350 12 2 two two NUM fcis-6350 12 3 - - PUNCT fcis-6350 12 4 stage stage NOUN fcis-6350 12 5 network network NOUN fcis-6350 12 6 has have VERB fcis-6350 12 7 high high ADJ fcis-6350 12 8 detection detection NOUN fcis-6350 12 9 accuracy accuracy NOUN fcis-6350 12 10 but but CCONJ fcis-6350 12 11 slow slow ADJ fcis-6350 12 12 detection detection NOUN fcis-6350 12 13 speed speed NOUN fcis-6350 12 14 , , PUNCT fcis-6350 12 15 and and CCONJ fcis-6350 12 16 the the DET fcis-6350 12 17 single single ADJ fcis-6350 12 18 - - PUNCT fcis-6350 12 19 stage stage NOUN fcis-6350 12 20 network network NOUN fcis-6350 12 21 has have VERB fcis-6350 12 22 fast fast ADJ fcis-6350 12 23 detection detection NOUN fcis-6350 12 24 speed speed NOUN fcis-6350 12 25 but but CCONJ fcis-6350 12 26 slightly slightly ADV fcis-6350 12 27 lower low ADJ fcis-6350 12 28 detection detection NOUN fcis-6350 12 29 accuracy accuracy NOUN fcis-6350 12 30 . . PUNCT fcis-6350 13 1 for for ADP fcis-6350 13 2 target target NOUN fcis-6350 13 3 detection detection NOUN fcis-6350 13 4 , , PUNCT fcis-6350 13 5 many many ADJ fcis-6350 13 6 scholars scholar NOUN fcis-6350 13 7 have have AUX fcis-6350 13 8 been be AUX fcis-6350 13 9 improving improve VERB fcis-6350 13 10 the the DET fcis-6350 13 11 capability capability NOUN fcis-6350 13 12 of of ADP fcis-6350 13 13 target target NOUN fcis-6350 13 14 detection detection NOUN fcis-6350 13 15 networks network NOUN fcis-6350 13 16 to to PART fcis-6350 13 17 achieve achieve VERB fcis-6350 13 18 good good ADJ fcis-6350 13 19 detection detection NOUN fcis-6350 13 20 results result NOUN fcis-6350 13 21 , , PUNCT fcis-6350 13 22 ren ren PROPN fcis-6350 13 23 et et PROPN fcis-6350 13 24 al al PROPN fcis-6350 13 25 improved improve VERB fcis-6350 13 26 the the DET fcis-6350 13 27 ssd ssd NOUN fcis-6350 13 28 algorithm algorithm NOUN fcis-6350 13 29 to to PART fcis-6350 13 30 enhance enhance VERB fcis-6350 13 31 the the DET fcis-6350 13 32 detection detection NOUN fcis-6350 13 33 capability capability NOUN fcis-6350 13 34 of of ADP fcis-6350 13 35 the the DET fcis-6350 13 36 network network NOUN fcis-6350 13 37 for for ADP fcis-6350 13 38 obscured obscured ADJ fcis-6350 13 39 objects object NOUN fcis-6350 13 40 or or CCONJ fcis-6350 13 41 small small ADJ fcis-6350 13 42 objects object NOUN fcis-6350 13 43 . . PUNCT fcis-6350 14 1 li li PROPN fcis-6350 14 2 et et PROPN fcis-6350 14 3 al al PROPN fcis-6350 14 4 proposed propose VERB fcis-6350 14 5 quality quality NOUN fcis-6350 14 6 focal focal ADJ fcis-6350 14 7 loss loss NOUN fcis-6350 14 8 to to PART fcis-6350 14 9 improve improve VERB fcis-6350 14 10 the the DET fcis-6350 14 11 error error NOUN fcis-6350 14 12 suppression suppression NOUN fcis-6350 14 13 that that PRON fcis-6350 14 14 occurs occur VERB fcis-6350 14 15 during during ADP fcis-6350 14 16 nms nms PROPN fcis-6350 14 17 operation operation NOUN fcis-6350 14 18 to to PART fcis-6350 14 19 improve improve VERB fcis-6350 14 20 the the DET fcis-6350 14 21 detection detection NOUN fcis-6350 14 22 accuracy accuracy NOUN fcis-6350 14 23 of of ADP fcis-6350 14 24 the the DET fcis-6350 14 25 network network NOUN fcis-6350 14 26 . . PUNCT fcis-6350 15 1 huang huang PROPN fcis-6350 15 2 et et PROPN fcis-6350 15 3 al al PROPN fcis-6350 15 4 designed design VERB fcis-6350 15 5 a a DET fcis-6350 15 6 variable variable ADJ fcis-6350 15 7 residual residual ADJ fcis-6350 15 8 convolutional convolutional ADJ fcis-6350 15 9 branching branching NOUN fcis-6350 15 10 module module NOUN fcis-6350 15 11 with with ADP fcis-6350 15 12 retinanet retinanet NOUN fcis-6350 15 13 as as ADP fcis-6350 15 14 the the DET fcis-6350 15 15 base base NOUN fcis-6350 15 16 framework framework NOUN fcis-6350 15 17 to to PART fcis-6350 15 18 improve improve VERB fcis-6350 15 19 the the DET fcis-6350 15 20 detection detection NOUN fcis-6350 15 21 accuracy accuracy NOUN fcis-6350 15 22 of of ADP fcis-6350 15 23 various various ADJ fcis-6350 15 24 types type NOUN fcis-6350 15 25 of of ADP fcis-6350 15 26 targets target NOUN fcis-6350 15 27 . . PUNCT fcis-6350 16 1 however however ADV fcis-6350 16 2 , , PUNCT fcis-6350 16 3 with with ADP fcis-6350 16 4 the the DET fcis-6350 16 5 improvement improvement NOUN fcis-6350 16 6 of of ADP fcis-6350 16 7 detection detection NOUN fcis-6350 16 8 capability capability NOUN fcis-6350 16 9 , , PUNCT fcis-6350 16 10 the the DET fcis-6350 16 11 network network NOUN fcis-6350 16 12 structure structure NOUN fcis-6350 16 13 becomes become VERB fcis-6350 16 14 more more ADV fcis-6350 16 15 and and CCONJ fcis-6350 16 16 more more ADV fcis-6350 16 17 complex complex ADJ fcis-6350 16 18 , , PUNCT fcis-6350 16 19 the the DET fcis-6350 16 20 number number NOUN fcis-6350 16 21 of of ADP fcis-6350 16 22 parameters parameter NOUN fcis-6350 16 23 , , PUNCT fcis-6350 16 24 and and CCONJ fcis-6350 16 25 the the DET fcis-6350 16 26 model model NOUN fcis-6350 16 27 size size NOUN fcis-6350 16 28 also also ADV fcis-6350 16 29 become become VERB fcis-6350 16 30 larger large ADJ fcis-6350 16 31 , , PUNCT fcis-6350 16 32 leading lead VERB fcis-6350 16 33 to to ADP fcis-6350 16 34 the the DET fcis-6350 16 35 decrease decrease NOUN fcis-6350 16 36 of of ADP fcis-6350 16 37 detection detection NOUN fcis-6350 16 38 speed speed NOUN fcis-6350 16 39 . . PUNCT fcis-6350 17 1 based base VERB fcis-6350 17 2 on on ADP fcis-6350 17 3 the the DET fcis-6350 17 4 above above ADJ fcis-6350 17 5 problems problem NOUN fcis-6350 17 6 , , PUNCT fcis-6350 17 7 this this DET fcis-6350 17 8 paper paper NOUN fcis-6350 17 9 performs perform VERB fcis-6350 17 10 light light NOUN fcis-6350 17 11 weighting weight VERB fcis-6350 17 12 on on ADP fcis-6350 17 13 the the DET fcis-6350 17 14 basis basis NOUN fcis-6350 17 15 of of ADP fcis-6350 17 16 yolov5s yolov5s PROPN fcis-6350 17 17 , , PUNCT fcis-6350 17 18 and and CCONJ fcis-6350 17 19 introduces introduce VERB fcis-6350 17 20 od od ADP fcis-6350 17 21 dynamic dynamic ADJ fcis-6350 17 22 convolution convolution NOUN fcis-6350 17 23 to to PART fcis-6350 17 24 improve improve VERB fcis-6350 17 25 the the DET fcis-6350 17 26 feature feature NOUN fcis-6350 17 27 extraction extraction NOUN fcis-6350 17 28 capability capability NOUN fcis-6350 17 29 of of ADP fcis-6350 17 30 the the DET fcis-6350 17 31 network network NOUN fcis-6350 17 32 by by ADP fcis-6350 17 33 in in ADP fcis-6350 17 34 order order NOUN fcis-6350 17 35 to to PART fcis-6350 17 36 compensate compensate VERB fcis-6350 17 37 for for ADP fcis-6350 17 38 the the DET fcis-6350 17 39 loss loss NOUN fcis-6350 17 40 of of ADP fcis-6350 17 41 model model NOUN fcis-6350 17 42 accuracy accuracy NOUN fcis-6350 17 43 after after ADP fcis-6350 17 44 light light ADJ fcis-6350 17 45 weighting weighting NOUN fcis-6350 17 46 . . PUNCT fcis-6350 18 1 the the DET fcis-6350 18 2 improved improved ADJ fcis-6350 18 3 network network NOUN fcis-6350 18 4 can can AUX fcis-6350 18 5 guarantee guarantee VERB fcis-6350 18 6 the the DET fcis-6350 18 7 detection detection NOUN fcis-6350 18 8 capability capability NOUN fcis-6350 18 9 on on ADP fcis-6350 18 10 the the DET fcis-6350 18 11 basis basis NOUN fcis-6350 18 12 of of ADP fcis-6350 18 13 lightweighting lightweighte VERB fcis-6350 18 14 . . PUNCT fcis-6350 19 1 it it PRON fcis-6350 19 2 is be AUX fcis-6350 19 3 beneficial beneficial ADJ fcis-6350 19 4 to to PART fcis-6350 19 5 be be AUX fcis-6350 19 6 deployed deploy VERB fcis-6350 19 7 on on ADP fcis-6350 19 8 in in ADP fcis-6350 19 9 - - PUNCT fcis-6350 19 10 vehicle vehicle NOUN fcis-6350 19 11 devices device NOUN fcis-6350 19 12 . . PUNCT fcis-6350 20 1 2 2 X fcis-6350 20 2 . . X fcis-6350 20 3 related relate VERB fcis-6350 20 4 work work NOUN fcis-6350 20 5 yolov5 yolov5 NOUN fcis-6350 20 6 algorithm algorithm PROPN fcis-6350 20 7 is be AUX fcis-6350 20 8 the the DET fcis-6350 20 9 latest late ADJ fcis-6350 20 10 version version NOUN fcis-6350 20 11 of of ADP fcis-6350 20 12 the the DET fcis-6350 20 13 current current ADJ fcis-6350 20 14 yolo yolo ADJ fcis-6350 20 15 series series PROPN fcis-6350 20 16 detection detection NOUN fcis-6350 20 17 algorithm algorithm NOUN fcis-6350 20 18 , , PUNCT fcis-6350 20 19 proposed propose VERB fcis-6350 20 20 by by ADP fcis-6350 20 21 ultralytics ultralytic NOUN fcis-6350 20 22 yolov5 yolov5 NOUN fcis-6350 20 23 main main ADJ fcis-6350 20 24 structure structure NOUN fcis-6350 20 25 is be AUX fcis-6350 20 26 shown show VERB fcis-6350 20 27 in in ADP fcis-6350 20 28 figure figure NOUN fcis-6350 20 29 1 1 NUM fcis-6350 20 30 , , PUNCT fcis-6350 20 31 the the DET fcis-6350 20 32 model model NOUN fcis-6350 20 33 consists consist VERB fcis-6350 20 34 of of ADP fcis-6350 20 35 four four NUM fcis-6350 20 36 parts part NOUN fcis-6350 20 37 : : PUNCT fcis-6350 20 38 input input NOUN fcis-6350 20 39 side side NOUN fcis-6350 20 40 ( ( PUNCT fcis-6350 20 41 input input NOUN fcis-6350 20 42 ) ) PUNCT fcis-6350 20 43 , , PUNCT fcis-6350 20 44 backbone backbone NOUN fcis-6350 20 45 network network NOUN fcis-6350 20 46 ( ( PUNCT fcis-6350 20 47 backbone backbone PROPN fcis-6350 20 48 ) ) PUNCT fcis-6350 20 49 , , PUNCT fcis-6350 20 50 multi multi ADJ fcis-6350 20 51 - - ADJ fcis-6350 20 52 scale scale ADJ fcis-6350 20 53 feature feature NOUN fcis-6350 20 54 fusion fusion NOUN fcis-6350 20 55 network network NOUN fcis-6350 20 56 ( ( PUNCT fcis-6350 20 57 neck neck NOUN fcis-6350 20 58 ) ) PUNCT fcis-6350 20 59 , , PUNCT fcis-6350 20 60 and and CCONJ fcis-6350 20 61 detection detection NOUN fcis-6350 20 62 network network NOUN fcis-6350 20 63 ( ( PUNCT fcis-6350 20 64 yolohead yolohead PROPN fcis-6350 20 65 ) ) PUNCT fcis-6350 20 66 . . PUNCT fcis-6350 21 1 figure figure NOUN fcis-6350 21 2 1 1 NUM fcis-6350 21 3 . . PUNCT fcis-6350 22 1 yolov5 yolov5 NOUN fcis-6350 22 2 network network NOUN fcis-6350 22 3 yolov5 yolov5 NOUN fcis-6350 22 4 input input NOUN fcis-6350 22 5 consists consist VERB fcis-6350 22 6 of of ADP fcis-6350 22 7 spp spp NOUN fcis-6350 22 8 structure structure NOUN fcis-6350 22 9 , , PUNCT fcis-6350 22 10 foucs foucs NOUN fcis-6350 22 11 structure structure NOUN fcis-6350 22 12 , , PUNCT fcis-6350 22 13 and and CCONJ fcis-6350 22 14 c3 c3 PROPN fcis-6350 22 15 structure structure NOUN fcis-6350 22 16 , , PUNCT fcis-6350 22 17 which which PRON fcis-6350 22 18 are be AUX fcis-6350 22 19 used use VERB fcis-6350 22 20 to to PART fcis-6350 22 21 extract extract VERB fcis-6350 22 22 the the DET fcis-6350 22 23 image image NOUN fcis-6350 22 24 network network NOUN fcis-6350 22 25 features.neck features.neck ADJ fcis-6350 22 26 side side NOUN fcis-6350 22 27 consists consist VERB fcis-6350 22 28 of of ADP fcis-6350 22 29 multi multi ADJ fcis-6350 22 30 - - ADJ fcis-6350 22 31 scale scale ADJ fcis-6350 22 32 pyramid pyramid NOUN fcis-6350 22 33 fpn+pan fpn+pan PROPN fcis-6350 22 34 structure structure NOUN fcis-6350 22 35 , , PUNCT fcis-6350 22 36 through through ADP fcis-6350 22 37 such such DET fcis-6350 22 38 a a DET fcis-6350 22 39 pyramid pyramid NOUN fcis-6350 22 40 structure structure NOUN fcis-6350 22 41 to to PART fcis-6350 22 42 increase increase VERB fcis-6350 22 43 the the DET fcis-6350 22 44 resolution resolution NOUN fcis-6350 22 45 of of ADP fcis-6350 22 46 the the DET fcis-6350 22 47 image image NOUN fcis-6350 22 48 , , PUNCT fcis-6350 22 49 so so SCONJ fcis-6350 22 50 that that SCONJ fcis-6350 22 51 the the DET fcis-6350 22 52 image image NOUN fcis-6350 22 53 can can AUX fcis-6350 22 54 be be AUX fcis-6350 22 55 divided divide VERB fcis-6350 22 56 into into ADP fcis-6350 22 57 more more ADJ fcis-6350 22 58 size size NOUN fcis-6350 22 59 pyramids pyramid NOUN fcis-6350 22 60 . . PUNCT fcis-6350 23 1 the the DET fcis-6350 23 2 more more ADJ fcis-6350 23 3 layers layer NOUN fcis-6350 23 4 of of ADP fcis-6350 23 5 the the DET fcis-6350 23 6 pyramid pyramid NOUN fcis-6350 23 7 , , PUNCT fcis-6350 23 8 the the DET fcis-6350 23 9 more more ADV fcis-6350 23 10 135 135 NUM fcis-6350 23 11 details detail NOUN fcis-6350 23 12 can can AUX fcis-6350 23 13 be be AUX fcis-6350 23 14 extracted extract VERB fcis-6350 23 15 . . PUNCT fcis-6350 24 1 in in ADP fcis-6350 24 2 addition addition NOUN fcis-6350 24 3 , , PUNCT fcis-6350 24 4 a a DET fcis-6350 24 5 batch batch NOUN fcis-6350 24 6 normalization normalization NOUN fcis-6350 24 7 layer layer NOUN fcis-6350 24 8 is be AUX fcis-6350 24 9 used use VERB fcis-6350 24 10 , , PUNCT fcis-6350 24 11 which which PRON fcis-6350 24 12 can can AUX fcis-6350 24 13 effectively effectively ADV fcis-6350 24 14 prevent prevent VERB fcis-6350 24 15 the the DET fcis-6350 24 16 occurrence occurrence NOUN fcis-6350 24 17 of of ADP fcis-6350 24 18 overfitting overfitting NOUN fcis-6350 24 19 . . PUNCT fcis-6350 25 1 the the DET fcis-6350 25 2 prediction prediction NOUN fcis-6350 25 3 side side NOUN fcis-6350 25 4 consists consist VERB fcis-6350 25 5 of of ADP fcis-6350 25 6 three three NUM fcis-6350 25 7 detection detection NOUN fcis-6350 25 8 heads head NOUN fcis-6350 25 9 , , PUNCT fcis-6350 25 10 corresponding correspond VERB fcis-6350 25 11 to to ADP fcis-6350 25 12 large large ADJ fcis-6350 25 13 , , PUNCT fcis-6350 25 14 medium medium ADJ fcis-6350 25 15 and and CCONJ fcis-6350 25 16 small small ADJ fcis-6350 25 17 size size NOUN fcis-6350 25 18 targets target NOUN fcis-6350 25 19 . . PUNCT fcis-6350 26 1 the the DET fcis-6350 26 2 yolov5 yolov5 NOUN fcis-6350 26 3 structure structure NOUN fcis-6350 26 4 is be AUX fcis-6350 26 5 shown show VERB fcis-6350 26 6 in in ADP fcis-6350 26 7 figure figure NOUN fcis-6350 26 8 1 1 NUM fcis-6350 26 9 . . NOUN fcis-6350 26 10 3 3 X fcis-6350 26 11 . . NOUN fcis-6350 26 12 method method NOUN fcis-6350 26 13 in in ADP fcis-6350 26 14 this this DET fcis-6350 26 15 paper paper NOUN fcis-6350 26 16 , , PUNCT fcis-6350 26 17 we we PRON fcis-6350 26 18 optimize optimize VERB fcis-6350 26 19 yolov5s yolov5s PROPN fcis-6350 26 20 from from ADP fcis-6350 26 21 two two NUM fcis-6350 26 22 aspects aspect NOUN fcis-6350 26 23 : : PUNCT fcis-6350 26 24 lightweight lightweight NOUN fcis-6350 26 25 and and CCONJ fcis-6350 26 26 attention attention NOUN fcis-6350 26 27 . . PUNCT fcis-6350 27 1 firstly firstly ADV fcis-6350 27 2 , , PUNCT fcis-6350 27 3 the the DET fcis-6350 27 4 yolov5s yolov5s PROPN fcis-6350 27 5 backbone backbone NOUN fcis-6350 27 6 network network NOUN fcis-6350 27 7 is be AUX fcis-6350 27 8 replaced replace VERB fcis-6350 27 9 with with ADP fcis-6350 27 10 efficientnetv2 efficientnetv2 NOUN fcis-6350 27 11 network network NOUN fcis-6350 27 12 to to PART fcis-6350 27 13 reduce reduce VERB fcis-6350 27 14 the the DET fcis-6350 27 15 computational computational ADJ fcis-6350 27 16 complexity complexity NOUN fcis-6350 27 17 and and CCONJ fcis-6350 27 18 make make VERB fcis-6350 27 19 the the DET fcis-6350 27 20 detection detection NOUN fcis-6350 27 21 model model VERB fcis-6350 27 22 more more ADV fcis-6350 27 23 lightweight lightweight ADJ fcis-6350 27 24 , , PUNCT fcis-6350 27 25 and and CCONJ fcis-6350 27 26 secondly secondly ADV fcis-6350 27 27 , , PUNCT fcis-6350 27 28 the the DET fcis-6350 27 29 conventional conventional ADJ fcis-6350 27 30 convolution convolution NOUN fcis-6350 27 31 is be AUX fcis-6350 27 32 replaced replace VERB fcis-6350 27 33 with with ADP fcis-6350 27 34 odconv odconv ADJ fcis-6350 27 35 dynamic dynamic ADJ fcis-6350 27 36 convolution convolution NOUN fcis-6350 27 37 in in ADP fcis-6350 27 38 the the DET fcis-6350 27 39 output output NOUN fcis-6350 27 40 layer layer NOUN fcis-6350 27 41 , , PUNCT fcis-6350 27 42 which which PRON fcis-6350 27 43 uses use VERB fcis-6350 27 44 a a DET fcis-6350 27 45 multidimensional multidimensional ADJ fcis-6350 27 46 attention attention NOUN fcis-6350 27 47 mechanism mechanism NOUN fcis-6350 27 48 to to PART fcis-6350 27 49 learn learn VERB fcis-6350 27 50 complementary complementary ADJ fcis-6350 27 51 attention attention NOUN fcis-6350 27 52 along along ADP fcis-6350 27 53 four four NUM fcis-6350 27 54 dimensions dimension NOUN fcis-6350 27 55 of of ADP fcis-6350 27 56 the the DET fcis-6350 27 57 kernel kernel PROPN fcis-6350 27 58 space space NOUN fcis-6350 27 59 through through ADP fcis-6350 27 60 a a DET fcis-6350 27 61 parallel parallel ADJ fcis-6350 27 62 strategy strategy NOUN fcis-6350 27 63 . . PUNCT fcis-6350 28 1 3.1 3.1 NUM fcis-6350 28 2 . . PUNCT fcis-6350 29 1 lightweight lightweight ADJ fcis-6350 29 2 efficientnetv2 efficientnetv2 PROPN fcis-6350 29 3 is be AUX fcis-6350 29 4 a a DET fcis-6350 29 5 lightweight lightweight ADJ fcis-6350 29 6 network network NOUN fcis-6350 29 7 that that PRON fcis-6350 29 8 combines combine VERB fcis-6350 29 9 training training NOUN fcis-6350 29 10 perceptual perceptual ADJ fcis-6350 29 11 neural neural ADJ fcis-6350 29 12 architecture architecture NOUN fcis-6350 29 13 search search NOUN fcis-6350 29 14 and and CCONJ fcis-6350 29 15 scaling scaling NOUN fcis-6350 29 16 , , PUNCT fcis-6350 29 17 which which PRON fcis-6350 29 18 better well ADJ fcis-6350 29 19 balances balance VERB fcis-6350 29 20 the the DET fcis-6350 29 21 relationship relationship NOUN fcis-6350 29 22 between between ADP fcis-6350 29 23 accuracy accuracy NOUN fcis-6350 29 24 , , PUNCT fcis-6350 29 25 training training NOUN fcis-6350 29 26 speed speed NOUN fcis-6350 29 27 and and CCONJ fcis-6350 29 28 number number NOUN fcis-6350 29 29 of of ADP fcis-6350 29 30 parameters.efficientnetv2 parameters.efficientnetv2 NOUN fcis-6350 29 31 - - PUNCT fcis-6350 29 32 s s NOUN fcis-6350 29 33 structure structure NOUN fcis-6350 29 34 table table NOUN fcis-6350 29 35 is be AUX fcis-6350 29 36 shown show VERB fcis-6350 29 37 in in ADP fcis-6350 29 38 table table NOUN fcis-6350 29 39 x. x. NOUN fcis-6350 29 40 table table NOUN fcis-6350 30 1 1 1 NUM fcis-6350 30 2 . . PUNCT fcis-6350 31 1 efficientnetv2s efficientnetv2s ADJ fcis-6350 31 2 structure structure NOUN fcis-6350 31 3 table table NOUN fcis-6350 31 4 stage stage NOUN fcis-6350 31 5 operator operator NOUN fcis-6350 31 6 stride stride NOUN fcis-6350 31 7 channel channel NOUN fcis-6350 31 8 layers layer NOUN fcis-6350 31 9 0 0 NUM fcis-6350 31 10 1 1 NUM fcis-6350 31 11 2 2 NUM fcis-6350 31 12 3 3 NUM fcis-6350 31 13 4 4 NUM fcis-6350 31 14 5 5 NUM fcis-6350 31 15 6 6 NUM fcis-6350 31 16 7 7 NUM fcis-6350 31 17 conv3x3 conv3x3 NUM fcis-6350 31 18 fused fuse VERB fcis-6350 31 19 - - PUNCT fcis-6350 31 20 mbconv1 mbconv1 NOUN fcis-6350 31 21 , , PUNCT fcis-6350 31 22 k3x3 k3x3 X fcis-6350 31 23 fused fuse VERB fcis-6350 31 24 - - PUNCT fcis-6350 31 25 mbconv4 mbconv4 NOUN fcis-6350 31 26 , , PUNCT fcis-6350 31 27 k3x3 k3x3 X fcis-6350 31 28 fused fuse VERB fcis-6350 31 29 - - PUNCT fcis-6350 31 30 mbconv4 mbconv4 NOUN fcis-6350 31 31 , , PUNCT fcis-6350 31 32 k3x3 k3x3 X fcis-6350 31 33 mbconv4,k3x3,se0.25 mbconv4,k3x3,se0.25 NOUN fcis-6350 31 34 mbconv6,k3x3,se0.25 mbconv6,k3x3,se0.25 PROPN fcis-6350 31 35 mbconv6,k3x3,se0.25 mbconv6,k3x3,se0.25 NOUN fcis-6350 31 36 conv1x1&pooling&fc conv1x1&pooling&fc NOUN fcis-6350 32 1 2 2 NUM fcis-6350 32 2 1 1 NUM fcis-6350 32 3 2 2 NUM fcis-6350 32 4 2 2 NUM fcis-6350 32 5 2 2 NUM fcis-6350 32 6 1 1 NUM fcis-6350 32 7 2 2 NUM fcis-6350 32 8 24 24 NUM fcis-6350 32 9 24 24 NUM fcis-6350 32 10 48 48 NUM fcis-6350 32 11 64 64 NUM fcis-6350 32 12 128 128 NUM fcis-6350 32 13 160 160 NUM fcis-6350 32 14 272 272 NUM fcis-6350 32 15 1792 1792 NUM fcis-6350 32 16 1 1 NUM fcis-6350 32 17 2 2 NUM fcis-6350 32 18 4 4 NUM fcis-6350 32 19 4 4 NUM fcis-6350 32 20 6 6 NUM fcis-6350 32 21 9 9 NUM fcis-6350 32 22 15 15 NUM fcis-6350 32 23 1 1 NUM fcis-6350 32 24 to to PART fcis-6350 32 25 address address VERB fcis-6350 32 26 the the DET fcis-6350 32 27 problem problem NOUN fcis-6350 32 28 of of ADP fcis-6350 32 29 slow slow ADJ fcis-6350 32 30 speed speed NOUN fcis-6350 32 31 due due ADP fcis-6350 32 32 to to ADP fcis-6350 32 33 dw dw NOUN fcis-6350 32 34 convolution convolution NOUN fcis-6350 32 35 in in ADP fcis-6350 32 36 the the DET fcis-6350 32 37 mbconv mbconv NOUN fcis-6350 32 38 module module NOUN fcis-6350 32 39 in in ADP fcis-6350 32 40 efficientnet efficientnet NOUN fcis-6350 32 41 , , PUNCT fcis-6350 32 42 the the DET fcis-6350 32 43 efficientnetv2 efficientnetv2 PROPN fcis-6350 32 44 network network NOUN fcis-6350 32 45 addresses address VERB fcis-6350 32 46 the the DET fcis-6350 32 47 problem problem NOUN fcis-6350 32 48 of of ADP fcis-6350 32 49 accuracy accuracy NOUN fcis-6350 32 50 degradation degradation NOUN fcis-6350 32 51 and and CCONJ fcis-6350 32 52 slower slow ADJ fcis-6350 32 53 training training NOUN fcis-6350 32 54 by by ADP fcis-6350 32 55 using use VERB fcis-6350 32 56 the the DET fcis-6350 32 57 fused fuse VERB fcis-6350 32 58 - - PUNCT fcis-6350 32 59 mbconv mbconv NOUN fcis-6350 32 60 structure structure NOUN fcis-6350 32 61 instead instead ADV fcis-6350 32 62 of of ADP fcis-6350 32 63 the the DET fcis-6350 32 64 mbconv mbconv NOUN fcis-6350 32 65 structure structure NOUN fcis-6350 32 66 in in ADP fcis-6350 32 67 the the DET fcis-6350 32 68 efficientnet efficientnet NOUN fcis-6350 32 69 network.the network.the DET fcis-6350 32 70 fusedmbconv fusedmbconv NOUN fcis-6350 32 71 replaces replace VERB fcis-6350 32 72 the the DET fcis-6350 32 73 1x1 1x1 NUM fcis-6350 32 74 normal normal ADJ fcis-6350 32 75 convolution convolution NOUN fcis-6350 32 76 and and CCONJ fcis-6350 32 77 3x3 3x3 NUM fcis-6350 32 78 deep deep ADJ fcis-6350 32 79 separable separable ADJ fcis-6350 32 80 convolution convolution NOUN fcis-6350 32 81 in in ADP fcis-6350 32 82 mbconv mbconv NOUN fcis-6350 32 83 by by ADP fcis-6350 32 84 3x3 3x3 NUM fcis-6350 32 85 normal normal ADJ fcis-6350 32 86 convolution convolution NOUN fcis-6350 32 87 , , PUNCT fcis-6350 32 88 and and CCONJ fcis-6350 32 89 then then ADV fcis-6350 32 90 uses use VERB fcis-6350 32 91 the the DET fcis-6350 32 92 nas nas NOUN fcis-6350 32 93 technique technique NOUN fcis-6350 32 94 to to PART fcis-6350 32 95 search search VERB fcis-6350 32 96 for for ADP fcis-6350 32 97 the the DET fcis-6350 32 98 best good ADJ fcis-6350 32 99 combined combine VERB fcis-6350 32 100 structure structure NOUN fcis-6350 32 101 , , PUNCT fcis-6350 32 102 the the DET fcis-6350 32 103 network network NOUN fcis-6350 32 104 structures structure NOUN fcis-6350 32 105 of of ADP fcis-6350 32 106 mbconv mbconv NOUN fcis-6350 32 107 and and CCONJ fcis-6350 32 108 fused fuse VERB fcis-6350 32 109 - - PUNCT fcis-6350 32 110 mbconv mbconv NOUN fcis-6350 32 111 are be AUX fcis-6350 32 112 shown show VERB fcis-6350 32 113 in in ADP fcis-6350 32 114 figure figure NOUN fcis-6350 32 115 2 2 NUM fcis-6350 32 116 , , PUNCT fcis-6350 32 117 figure figure VERB fcis-6350 32 118 3 3 NUM fcis-6350 32 119 figure figure NOUN fcis-6350 32 120 2 2 NUM fcis-6350 32 121 . . NOUN fcis-6350 32 122 mbconv mbconv NOUN fcis-6350 32 123 module module NOUN fcis-6350 32 124 figure figure NOUN fcis-6350 32 125 3 3 NUM fcis-6350 32 126 . . PUNCT fcis-6350 32 127 fused fuse VERB fcis-6350 32 128 - - PUNCT fcis-6350 32 129 mbconv mbconv NOUN fcis-6350 32 130 module module NOUN fcis-6350 32 131 the the DET fcis-6350 32 132 structure structure NOUN fcis-6350 32 133 of of ADP fcis-6350 32 134 the the DET fcis-6350 32 135 yolov5s yolov5s PROPN fcis-6350 32 136 backbone backbone NOUN fcis-6350 32 137 network network NOUN fcis-6350 32 138 after after ADP fcis-6350 32 139 replacing replace VERB fcis-6350 32 140 it it PRON fcis-6350 32 141 with with ADP fcis-6350 32 142 efficientnetv2 efficientnetv2 PROPN fcis-6350 32 143 is be AUX fcis-6350 32 144 shown show VERB fcis-6350 32 145 in in ADP fcis-6350 32 146 figure4 figure4 PROPN fcis-6350 32 147 . . PUNCT fcis-6350 33 1 figure figure NOUN fcis-6350 33 2 4 4 NUM fcis-6350 33 3 . . PUNCT fcis-6350 34 1 yolov5sefficientnetv2 yolov5sefficientnetv2 PROPN fcis-6350 34 2 structure structure PROPN fcis-6350 34 3 3.2 3.2 NUM fcis-6350 34 4 . . PUNCT fcis-6350 35 1 odconv odconv ADJ fcis-6350 35 2 odconv odconv ADJ fcis-6350 35 3 convolution convolution NOUN fcis-6350 35 4 takes take VERB fcis-6350 35 5 into into ADP fcis-6350 35 6 account account NOUN fcis-6350 35 7 the the DET fcis-6350 35 8 dynamics dynamic NOUN fcis-6350 35 9 of of ADP fcis-6350 35 10 the the DET fcis-6350 35 11 null null ADJ fcis-6350 35 12 field field NOUN fcis-6350 35 13 , , PUNCT fcis-6350 35 14 input input NOUN fcis-6350 35 15 channel channel NOUN fcis-6350 35 16 , , PUNCT fcis-6350 35 17 output output NOUN fcis-6350 35 18 channel channel NOUN fcis-6350 35 19 and and CCONJ fcis-6350 35 20 other other ADJ fcis-6350 35 21 dimensions dimension NOUN fcis-6350 35 22 , , PUNCT fcis-6350 35 23 so so SCONJ fcis-6350 35 24 it it PRON fcis-6350 35 25 is be AUX fcis-6350 35 26 called call VERB fcis-6350 35 27 full full ADJ fcis-6350 35 28 - - PUNCT fcis-6350 35 29 dimensional dimensional ADJ fcis-6350 35 30 dynamic dynamic ADJ fcis-6350 35 31 convolution convolution NOUN fcis-6350 35 32 . . PUNCT fcis-6350 36 1 odconv odconv PROPN fcis-6350 36 2 adopts adopt VERB fcis-6350 36 3 multidimensional multidimensional ADJ fcis-6350 36 4 attention attention NOUN fcis-6350 36 5 mechanism mechanism NOUN fcis-6350 36 6 to to PART fcis-6350 36 7 learn learn VERB fcis-6350 36 8 complementary complementary ADJ fcis-6350 36 9 attention attention NOUN fcis-6350 36 10 in in ADP fcis-6350 36 11 four four NUM fcis-6350 36 12 dimensions dimension NOUN fcis-6350 36 13 of of ADP fcis-6350 36 14 convolutional convolutional ADJ fcis-6350 36 15 kernel kernel NOUN fcis-6350 36 16 : : PUNCT fcis-6350 36 17 input input NOUN fcis-6350 36 18 channel channel PROPN fcis-6350 36 19 , , PUNCT fcis-6350 36 20 output output NOUN fcis-6350 36 21 channel channel PROPN fcis-6350 36 22 , , PUNCT fcis-6350 36 23 kernel kernel NOUN fcis-6350 36 24 space space NOUN fcis-6350 36 25 and and CCONJ fcis-6350 36 26 number number NOUN fcis-6350 36 27 of of ADP fcis-6350 36 28 kernels kernel NOUN fcis-6350 36 29 through through ADP fcis-6350 36 30 parallel parallel ADJ fcis-6350 36 31 strategy strategy NOUN fcis-6350 36 32 , , PUNCT fcis-6350 36 33 which which PRON fcis-6350 36 34 enhances enhance VERB fcis-6350 36 35 the the DET fcis-6350 36 36 feature feature NOUN fcis-6350 36 37 extraction extraction NOUN fcis-6350 36 38 ability ability NOUN fcis-6350 36 39 of of ADP fcis-6350 36 40 convolution convolution NOUN fcis-6350 36 41 in in ADP fcis-6350 36 42 all all DET fcis-6350 36 43 aspects aspect NOUN fcis-6350 36 44 . . PUNCT fcis-6350 37 1 the the DET fcis-6350 37 2 structure structure NOUN fcis-6350 37 3 of of ADP fcis-6350 37 4 odconv odconv ADJ fcis-6350 37 5 comparison comparison NOUN fcis-6350 37 6 is be AUX fcis-6350 37 7 shown show VERB fcis-6350 37 8 in in ADP fcis-6350 37 9 figure figure NOUN fcis-6350 37 10 5 5 NUM fcis-6350 37 11 . . PUNCT fcis-6350 37 12 figure figure VERB fcis-6350 37 13 5 5 NUM fcis-6350 37 14 . . PUNCT fcis-6350 37 15 dyconv dyconv NOUN fcis-6350 37 16 and and CCONJ fcis-6350 37 17 odconv odconv ADJ fcis-6350 37 18 structure structure NOUN fcis-6350 37 19 comparison comparison NOUN fcis-6350 37 20 diagram diagram VERB fcis-6350 37 21 the the DET fcis-6350 37 22 above above ADJ fcis-6350 37 23 figure figure NOUN fcis-6350 37 24 shows show VERB fcis-6350 37 25 a a DET fcis-6350 37 26 schematic schematic ADJ fcis-6350 37 27 comparison comparison NOUN fcis-6350 37 28 of of ADP fcis-6350 37 29 the the DET fcis-6350 37 30 structures structure NOUN fcis-6350 37 31 of of ADP fcis-6350 37 32 dyconv dyconv NOUN fcis-6350 37 33 ( ( PUNCT fcis-6350 37 34 a),odconv a),odconv NOUN fcis-6350 37 35 ( ( PUNCT fcis-6350 37 36 b b NOUN fcis-6350 37 37 ) ) PUNCT fcis-6350 37 38 . . PUNCT fcis-6350 38 1 unlike unlike ADP fcis-6350 38 2 condconv condconv NOUN fcis-6350 38 3 and and CCONJ fcis-6350 38 4 dyconv dyconv NOUN fcis-6350 38 5 , , PUNCT fcis-6350 38 6 which which PRON fcis-6350 38 7 compute compute VERB fcis-6350 38 8 a a DET fcis-6350 38 9 single single ADJ fcis-6350 38 10 attention attention NOUN fcis-6350 38 11 scalar scalar NOUN fcis-6350 38 12 , , PUNCT fcis-6350 38 13 for for SCONJ fcis-6350 38 14 the the DET fcis-6350 38 15 convolution convolution NOUN fcis-6350 38 16 kernel kernel PROPN fcis-6350 38 17 odconv odconv PROPN fcis-6350 38 18 utilizes utilize VERB fcis-6350 38 19 a a DET fcis-6350 38 20 new new ADJ fcis-6350 38 21 multidimensional multidimensional ADJ fcis-6350 38 22 attention attention NOUN fcis-6350 38 23 mechanism mechanism NOUN fcis-6350 38 24 that that PRON fcis-6350 38 25 computes compute VERB fcis-6350 38 26 four four NUM fcis-6350 38 27 types type NOUN fcis-6350 38 28 of of ADP fcis-6350 38 29 attention attention NOUN fcis-6350 38 30 in in ADP fcis-6350 38 31 parallel parallel NOUN fcis-6350 38 32 along along ADP fcis-6350 38 33 all all DET fcis-6350 38 34 four four NUM fcis-6350 38 35 dimensions dimension NOUN fcis-6350 38 36 of of ADP fcis-6350 38 37 the the DET fcis-6350 38 38 kernel kernel PROPN fcis-6350 38 39 space space NOUN fcis-6350 38 40 . . PUNCT fcis-6350 39 1 the the DET fcis-6350 39 2 four four NUM fcis-6350 39 3 types type NOUN fcis-6350 39 4 of of ADP fcis-6350 39 5 attention attention NOUN fcis-6350 39 6 convolution convolution NOUN fcis-6350 39 7 are be AUX fcis-6350 39 8 illustrated illustrate VERB fcis-6350 39 9 in in ADP fcis-6350 39 10 figure figure NOUN fcis-6350 39 11 6 6 NUM fcis-6350 39 12 figure figure NOUN fcis-6350 39 13 6 6 NUM fcis-6350 39 14 . . PUNCT fcis-6350 39 15 odconv odconv PROPN fcis-6350 39 16 illustration illustration NOUN fcis-6350 39 17 diagram diagram NOUN fcis-6350 39 18 ( ( PUNCT fcis-6350 39 19 a a NOUN fcis-6350 39 20 ) ) PUNCT fcis-6350 39 21 , , PUNCT fcis-6350 39 22 ( ( PUNCT fcis-6350 39 23 b b X fcis-6350 39 24 ) ) PUNCT fcis-6350 39 25 ( ( PUNCT fcis-6350 39 26 c c X fcis-6350 39 27 ) ) PUNCT fcis-6350 39 28 ( ( PUNCT fcis-6350 39 29 d d X fcis-6350 39 30 ) ) PUNCT fcis-6350 39 31 represent represent VERB fcis-6350 39 32 the the DET fcis-6350 39 33 position position NOUN fcis-6350 39 34 multiplication multiplication NOUN fcis-6350 39 35 along along ADP fcis-6350 39 36 the the DET fcis-6350 39 37 spatial spatial ADJ fcis-6350 39 38 dimension dimension NOUN fcis-6350 39 39 , , PUNCT fcis-6350 39 40 the the DET fcis-6350 39 41 channel channel NOUN fcis-6350 39 42 multiplication multiplication NOUN fcis-6350 39 43 along along ADP fcis-6350 39 44 the the DET fcis-6350 39 45 input input NOUN fcis-6350 39 46 channel channel NOUN fcis-6350 39 47 dimension dimension NOUN fcis-6350 39 48 , , PUNCT fcis-6350 39 49 the the DET fcis-6350 39 50 filter filter NOUN fcis-6350 39 51 multiplication multiplication NOUN fcis-6350 39 52 along along ADP fcis-6350 39 53 the the DET fcis-6350 39 54 output output NOUN fcis-6350 39 55 channel channel NOUN fcis-6350 39 56 dimension dimension NOUN fcis-6350 39 57 , , PUNCT fcis-6350 39 58 and and CCONJ fcis-6350 39 59 the the DET fcis-6350 39 60 kernel kernel PROPN fcis-6350 39 61 multiplication multiplication NOUN fcis-6350 39 62 along along ADP fcis-6350 39 63 the the DET fcis-6350 39 64 kernel kernel PROPN fcis-6350 39 65 dimension dimension NOUN fcis-6350 39 66 of of ADP fcis-6350 39 67 the the DET fcis-6350 39 68 convolution convolution NOUN fcis-6350 39 69 kernel kernel PROPN fcis-6350 39 70 space space NOUN fcis-6350 39 71 , , PUNCT fcis-6350 39 72 136 136 NUM fcis-6350 39 73 respectively respectively ADV fcis-6350 39 74 . . PUNCT fcis-6350 40 1 these these DET fcis-6350 40 2 four four NUM fcis-6350 40 3 types type NOUN fcis-6350 40 4 of of ADP fcis-6350 40 5 attention attention NOUN fcis-6350 40 6 are be AUX fcis-6350 40 7 complementary complementary ADJ fcis-6350 40 8 , , PUNCT fcis-6350 40 9 and and CCONJ fcis-6350 40 10 by by ADP fcis-6350 40 11 progressively progressively ADV fcis-6350 40 12 multiplying multiply VERB fcis-6350 40 13 the the DET fcis-6350 40 14 convolution convolution NOUN fcis-6350 40 15 with with ADP fcis-6350 40 16 different different ADJ fcis-6350 40 17 attentions attention NOUN fcis-6350 40 18 along along ADP fcis-6350 40 19 the the DET fcis-6350 40 20 position position NOUN fcis-6350 40 21 , , PUNCT fcis-6350 40 22 channel channel NOUN fcis-6350 40 23 , , PUNCT fcis-6350 40 24 filter filter NOUN fcis-6350 40 25 , , PUNCT fcis-6350 40 26 and and CCONJ fcis-6350 40 27 kernel kernel PROPN fcis-6350 40 28 dimensions dimension NOUN fcis-6350 40 29 will will AUX fcis-6350 40 30 make make VERB fcis-6350 40 31 the the DET fcis-6350 40 32 convolution convolution NOUN fcis-6350 40 33 operation operation NOUN fcis-6350 40 34 differ differ VERB fcis-6350 40 35 for for ADP fcis-6350 40 36 each each DET fcis-6350 40 37 dimension dimension NOUN fcis-6350 40 38 of of ADP fcis-6350 40 39 the the DET fcis-6350 40 40 input input NOUN fcis-6350 40 41 , , PUNCT fcis-6350 40 42 providing provide VERB fcis-6350 40 43 better well ADJ fcis-6350 40 44 performance performance NOUN fcis-6350 40 45 to to PART fcis-6350 40 46 capture capture VERB fcis-6350 40 47 rich rich ADJ fcis-6350 40 48 contextual contextual ADJ fcis-6350 40 49 information information NOUN fcis-6350 40 50 . . PUNCT fcis-6350 41 1 therefore therefore ADV fcis-6350 41 2 , , PUNCT fcis-6350 41 3 odconv odconv NOUN fcis-6350 41 4 can can AUX fcis-6350 41 5 significantly significantly ADV fcis-6350 41 6 improve improve VERB fcis-6350 41 7 the the DET fcis-6350 41 8 feature feature NOUN fcis-6350 41 9 extraction extraction NOUN fcis-6350 41 10 capability capability NOUN fcis-6350 41 11 of of ADP fcis-6350 41 12 convolution convolution NOUN fcis-6350 41 13 . . PUNCT fcis-6350 42 1 the the DET fcis-6350 42 2 structure structure NOUN fcis-6350 42 3 of of ADP fcis-6350 42 4 the the DET fcis-6350 42 5 network network NOUN fcis-6350 42 6 after after ADP fcis-6350 42 7 adding add VERB fcis-6350 42 8 dynamic dynamic ADJ fcis-6350 42 9 od od NOUN fcis-6350 42 10 convolution convolution NOUN fcis-6350 42 11 is be AUX fcis-6350 42 12 shown show VERB fcis-6350 42 13 in in ADP fcis-6350 42 14 figure figure NOUN fcis-6350 42 15 7 7 NUM fcis-6350 42 16 . . PUNCT fcis-6350 42 17 figure figure NOUN fcis-6350 42 18 7 7 NUM fcis-6350 42 19 . . PUNCT fcis-6350 43 1 yolov5sefficientnetv2 yolov5sefficientnetv2 PROPN fcis-6350 43 2 - - PUNCT fcis-6350 43 3 od od PROPN fcis-6350 43 4 structure structure NOUN fcis-6350 43 5 diagram diagram NOUN fcis-6350 43 6 3.3 3.3 NUM fcis-6350 43 7 . . PUNCT fcis-6350 44 1 dataset dataset VERB fcis-6350 44 2 the the DET fcis-6350 44 3 kitti kitti PROPN fcis-6350 44 4 dataset dataset PROPN fcis-6350 44 5 , , PUNCT fcis-6350 44 6 co co VERB fcis-6350 44 7 - - VERB fcis-6350 44 8 founded found VERB fcis-6350 44 9 by by ADP fcis-6350 44 10 karlsruhe karlsruhe PROPN fcis-6350 44 11 institute institute PROPN fcis-6350 44 12 of of ADP fcis-6350 44 13 technology technology PROPN fcis-6350 44 14 and and CCONJ fcis-6350 44 15 toyota toyota PROPN fcis-6350 44 16 institute institute PROPN fcis-6350 44 17 of of ADP fcis-6350 44 18 technology technology PROPN fcis-6350 44 19 , , PUNCT fcis-6350 44 20 is be AUX fcis-6350 44 21 the the DET fcis-6350 44 22 largest large ADJ fcis-6350 44 23 international international ADJ fcis-6350 44 24 dataset dataset NOUN fcis-6350 44 25 for for ADP fcis-6350 44 26 evaluating evaluate VERB fcis-6350 44 27 computer computer NOUN fcis-6350 44 28 vision vision NOUN fcis-6350 44 29 algorithms algorithm NOUN fcis-6350 44 30 in in ADP fcis-6350 44 31 autonomous autonomous ADJ fcis-6350 44 32 driving driving NOUN fcis-6350 44 33 scenarios scenario NOUN fcis-6350 44 34 . . PUNCT fcis-6350 45 1 the the DET fcis-6350 45 2 kitti kitti PROPN fcis-6350 45 3 includes include VERB fcis-6350 45 4 real real ADJ fcis-6350 45 5 image image NOUN fcis-6350 45 6 data datum NOUN fcis-6350 45 7 from from ADP fcis-6350 45 8 urban urban ADJ fcis-6350 45 9 , , PUNCT fcis-6350 45 10 rural rural ADJ fcis-6350 45 11 , , PUNCT fcis-6350 45 12 and and CCONJ fcis-6350 45 13 highway highway NOUN fcis-6350 45 14 scenes scene NOUN fcis-6350 45 15 , , PUNCT fcis-6350 45 16 with with ADP fcis-6350 45 17 up up ADP fcis-6350 45 18 to to PART fcis-6350 45 19 15 15 NUM fcis-6350 45 20 vehicles vehicle NOUN fcis-6350 45 21 and and CCONJ fcis-6350 45 22 30 30 NUM fcis-6350 45 23 pedestrians pedestrian NOUN fcis-6350 45 24 per per ADP fcis-6350 45 25 image image NOUN fcis-6350 45 26 , , PUNCT fcis-6350 45 27 and and CCONJ fcis-6350 45 28 various various ADJ fcis-6350 45 29 levels level NOUN fcis-6350 45 30 of of ADP fcis-6350 45 31 occlusion occlusion NOUN fcis-6350 45 32 and and CCONJ fcis-6350 45 33 truncation truncation NOUN fcis-6350 45 34 . . PUNCT fcis-6350 46 1 the the DET fcis-6350 46 2 original original ADJ fcis-6350 46 3 dataset dataset NOUN fcis-6350 46 4 is be AUX fcis-6350 46 5 classified classify VERB fcis-6350 46 6 and and CCONJ fcis-6350 46 7 subdivided subdivide VERB fcis-6350 46 8 into into ADP fcis-6350 46 9 car car NOUN fcis-6350 46 10 , , PUNCT fcis-6350 46 11 van van NOUN fcis-6350 46 12 , , PUNCT fcis-6350 46 13 truck truck NOUN fcis-6350 46 14 , , PUNCT fcis-6350 46 15 pedestrian pedestrian NOUN fcis-6350 46 16 , , PUNCT fcis-6350 46 17 pedestrian(sitting pedestrian(sitting PROPN fcis-6350 46 18 ) ) PUNCT fcis-6350 46 19 , , PUNCT fcis-6350 46 20 cyclist cyclist NOUN fcis-6350 46 21 , , PUNCT fcis-6350 46 22 tram tram PROPN fcis-6350 46 23 , , PUNCT fcis-6350 46 24 and and CCONJ fcis-6350 46 25 misc misc PROPN fcis-6350 46 26 . . PUNCT fcis-6350 47 1 in in ADP fcis-6350 47 2 this this DET fcis-6350 47 3 paper paper NOUN fcis-6350 47 4 , , PUNCT fcis-6350 47 5 the the DET fcis-6350 47 6 original original ADJ fcis-6350 47 7 kitti kitti PROPN fcis-6350 47 8 dataset dataset VERB fcis-6350 47 9 consists consist NOUN fcis-6350 47 10 of of ADP fcis-6350 47 11 car car NOUN fcis-6350 47 12 , , PUNCT fcis-6350 47 13 truck truck NOUN fcis-6350 47 14 , , PUNCT fcis-6350 47 15 van van NOUN fcis-6350 47 16 as as ADP fcis-6350 47 17 car car NOUN fcis-6350 47 18 , , PUNCT fcis-6350 47 19 pedestrian pedestrian NOUN fcis-6350 47 20 , , PUNCT fcis-6350 47 21 person_sitting person_sitte VERB fcis-6350 47 22 as as ADP fcis-6350 47 23 person person NOUN fcis-6350 47 24 , , PUNCT fcis-6350 47 25 and and CCONJ fcis-6350 47 26 cyclist cyclist NOUN fcis-6350 47 27 as as ADP fcis-6350 47 28 rider rider NOUN fcis-6350 47 29 , , PUNCT fcis-6350 47 30 and and CCONJ fcis-6350 47 31 the the DET fcis-6350 47 32 final final ADJ fcis-6350 47 33 kitti kitti PROPN fcis-6350 47 34 dataset dataset VERB fcis-6350 47 35 consists consist NOUN fcis-6350 47 36 of of ADP fcis-6350 47 37 car car NOUN fcis-6350 47 38 , , PUNCT fcis-6350 47 39 person person NOUN fcis-6350 47 40 , , PUNCT fcis-6350 47 41 and and CCONJ fcis-6350 47 42 rider rider NOUN fcis-6350 47 43 . . PUNCT fcis-6350 48 1 these these DET fcis-6350 48 2 three three NUM fcis-6350 48 3 categories category NOUN fcis-6350 48 4 . . PUNCT fcis-6350 49 1 3.4 3.4 NUM fcis-6350 49 2 . . PUNCT fcis-6350 49 3 configuration configuration NOUN fcis-6350 49 4 the the DET fcis-6350 49 5 operating operating NOUN fcis-6350 49 6 system system NOUN fcis-6350 49 7 of of ADP fcis-6350 49 8 the the DET fcis-6350 49 9 experimental experimental ADJ fcis-6350 49 10 platform platform NOUN fcis-6350 49 11 in in ADP fcis-6350 49 12 this this DET fcis-6350 49 13 paper paper NOUN fcis-6350 49 14 is be AUX fcis-6350 49 15 window10 window10 NOUN fcis-6350 49 16 , , PUNCT fcis-6350 49 17 the the DET fcis-6350 49 18 hardware hardware NOUN fcis-6350 49 19 configuration configuration NOUN fcis-6350 49 20 is be AUX fcis-6350 49 21 intel(r intel(r NOUN fcis-6350 49 22 ) ) PUNCT fcis-6350 49 23 core core NOUN fcis-6350 49 24 ( ( PUNCT fcis-6350 49 25 tm tm NOUN fcis-6350 49 26 ) ) PUNCT fcis-6350 49 27 i7 i7 NOUN fcis-6350 49 28 - - PUNCT fcis-6350 49 29 10875h 10875h NUM fcis-6350 49 30 2.30ghz 2.30ghz NUM fcis-6350 49 31 processor processor NOUN fcis-6350 49 32 , , PUNCT fcis-6350 49 33 geforcertx2060 geforcertx2060 NOUN fcis-6350 49 34 graphics graphic NOUN fcis-6350 49 35 card card NOUN fcis-6350 49 36 with with ADP fcis-6350 49 37 6 6 NUM fcis-6350 49 38 gb gb NOUN fcis-6350 49 39 of of ADP fcis-6350 49 40 video video NOUN fcis-6350 49 41 memory memory NOUN fcis-6350 49 42 size size NOUN fcis-6350 49 43 , , PUNCT fcis-6350 49 44 and and CCONJ fcis-6350 49 45 the the DET fcis-6350 49 46 software software NOUN fcis-6350 49 47 environment environment NOUN fcis-6350 49 48 equipped equip VERB fcis-6350 49 49 with with ADP fcis-6350 49 50 anaconda anaconda PROPN fcis-6350 49 51 , , PUNCT fcis-6350 49 52 cuda11.4.0 cuda11.4.0 NOUN fcis-6350 49 53 , , PUNCT fcis-6350 49 54 using use VERB fcis-6350 49 55 pytorch1.7.1 pytorch1.7.1 PRON fcis-6350 49 56 deep deep ADJ fcis-6350 49 57 learning learning NOUN fcis-6350 49 58 and and CCONJ fcis-6350 49 59 yolov5 yolov5 NOUN fcis-6350 49 60 target target NOUN fcis-6350 49 61 detection detection NOUN fcis-6350 49 62 framework framework NOUN fcis-6350 49 63 . . PUNCT fcis-6350 50 1 3.5 3.5 NUM fcis-6350 50 2 . . PUNCT fcis-6350 50 3 evaluation evaluation NOUN fcis-6350 50 4 indicators indicator NOUN fcis-6350 50 5 average average ADJ fcis-6350 50 6 precision precision NOUN fcis-6350 50 7 ( ( PUNCT fcis-6350 50 8 ap ap PROPN fcis-6350 50 9 ): ): PUNCT fcis-6350 50 10 the the DET fcis-6350 50 11 ap ap PROPN fcis-6350 50 12 value value NOUN fcis-6350 50 13 is be AUX fcis-6350 50 14 the the DET fcis-6350 50 15 area area NOUN fcis-6350 50 16 enclosed enclose VERB fcis-6350 50 17 by by ADP fcis-6350 50 18 the the DET fcis-6350 50 19 p p PROPN fcis-6350 50 20 - - PUNCT fcis-6350 50 21 r r NOUN fcis-6350 50 22 curve curve NOUN fcis-6350 50 23 ( ( PUNCT fcis-6350 50 24 consisting consist VERB fcis-6350 50 25 of of ADP fcis-6350 50 26 the the DET fcis-6350 50 27 values value NOUN fcis-6350 50 28 of of ADP fcis-6350 50 29 precision precision NOUN fcis-6350 50 30 p p NOUN fcis-6350 50 31 and and CCONJ fcis-6350 50 32 recall recall NOUN fcis-6350 50 33 r r NOUN fcis-6350 50 34 at at ADP fcis-6350 50 35 different different ADJ fcis-6350 50 36 confidence confidence NOUN fcis-6350 50 37 thresholds threshold NOUN fcis-6350 50 38 ) ) PUNCT fcis-6350 50 39 and and CCONJ fcis-6350 50 40 the the DET fcis-6350 50 41 coordinate coordinate NOUN fcis-6350 50 42 axes axis NOUN fcis-6350 50 43 , , PUNCT fcis-6350 50 44 and and CCONJ fcis-6350 50 45 the the DET fcis-6350 50 46 larger large ADJ fcis-6350 50 47 the the DET fcis-6350 50 48 area area NOUN fcis-6350 50 49 enclosed enclose VERB fcis-6350 50 50 by by ADP fcis-6350 50 51 the the DET fcis-6350 50 52 p p PROPN fcis-6350 50 53 - - PUNCT fcis-6350 50 54 r r NOUN fcis-6350 50 55 curve curve NOUN fcis-6350 50 56 , , PUNCT fcis-6350 50 57 the the PRON fcis-6350 50 58 better well ADJ fcis-6350 50 59 the the DET fcis-6350 50 60 detection detection NOUN fcis-6350 50 61 performance performance NOUN fcis-6350 50 62 of of ADP fcis-6350 50 63 the the DET fcis-6350 50 64 network network NOUN fcis-6350 50 65 model model NOUN fcis-6350 50 66 . . PUNCT fcis-6350 51 1 ๐ด๐‘ƒ ๐ด๐‘ƒ PROPN fcis-6350 51 2 = = SYM fcis-6350 51 3 โˆซ โˆซ NOUN fcis-6350 51 4 ๐‘ƒ(๐‘…)๐‘‘๐‘… ๐‘ƒ(๐‘…)๐‘‘๐‘… NOUN fcis-6350 51 5 1 1 NUM fcis-6350 51 6 0 0 NUM fcis-6350 51 7 ( ( PUNCT fcis-6350 51 8 1 1 X fcis-6350 51 9 ) ) PUNCT fcis-6350 51 10 mean mean ADJ fcis-6350 51 11 average average ADJ fcis-6350 51 12 precison precison NOUN fcis-6350 51 13 ( ( PUNCT fcis-6350 51 14 map map NOUN fcis-6350 51 15 ): ): PUNCT fcis-6350 51 16 indicates indicate VERB fcis-6350 51 17 that that SCONJ fcis-6350 51 18 the the DET fcis-6350 51 19 ap ap PROPN fcis-6350 51 20 of of ADP fcis-6350 51 21 multiple multiple ADJ fcis-6350 51 22 categories category NOUN fcis-6350 51 23 is be AUX fcis-6350 51 24 averaged average VERB fcis-6350 51 25 and and CCONJ fcis-6350 51 26 calculated calculate VERB fcis-6350 51 27 by by ADP fcis-6350 51 28 the the DET fcis-6350 51 29 following follow VERB fcis-6350 51 30 formula formula NOUN fcis-6350 51 31 : : PUNCT fcis-6350 51 32 ๐‘š๐‘Ž๐‘ ๐‘š๐‘Ž๐‘ X fcis-6350 51 33 = = PUNCT fcis-6350 51 34 โˆ‘ โˆ‘ PUNCT fcis-6350 51 35 ๐ด๐‘ƒ๐‘– ๐ด๐‘ƒ๐‘– PROPN fcis-6350 51 36 ๐‘› ๐‘› PROPN fcis-6350 51 37 ๐‘–=1 ๐‘–=1 PUNCT fcis-6350 51 38 ๐‘› ๐‘› PROPN fcis-6350 51 39 ( ( PUNCT fcis-6350 51 40 2 2 NUM fcis-6350 51 41 ) ) PUNCT fcis-6350 51 42 n n NOUN fcis-6350 51 43 denotes denote VERB fcis-6350 51 44 the the DET fcis-6350 51 45 number number NOUN fcis-6350 51 46 of of ADP fcis-6350 51 47 categories category NOUN fcis-6350 51 48 . . PUNCT fcis-6350 52 1 detection detection NOUN fcis-6350 52 2 speed speed NOUN fcis-6350 52 3 ( ( PUNCT fcis-6350 52 4 fps fps PROPN fcis-6350 52 5 ): ): PUNCT fcis-6350 52 6 reflects reflect VERB fcis-6350 52 7 the the DET fcis-6350 52 8 real real ADJ fcis-6350 52 9 - - PUNCT fcis-6350 52 10 time time NOUN fcis-6350 52 11 nature nature NOUN fcis-6350 52 12 of of ADP fcis-6350 52 13 the the DET fcis-6350 52 14 algorithm algorithm NOUN fcis-6350 52 15 , , PUNCT fcis-6350 52 16 and and CCONJ fcis-6350 52 17 being be AUX fcis-6350 52 18 able able ADJ fcis-6350 52 19 to to PART fcis-6350 52 20 detect detect VERB fcis-6350 52 21 obstacle obstacle NOUN fcis-6350 52 22 targets target NOUN fcis-6350 52 23 in in ADP fcis-6350 52 24 realtime realtime NOUN fcis-6350 52 25 is be AUX fcis-6350 52 26 a a DET fcis-6350 52 27 requirement requirement NOUN fcis-6350 52 28 for for ADP fcis-6350 52 29 self self NOUN fcis-6350 52 30 - - PUNCT fcis-6350 52 31 driving drive VERB fcis-6350 52 32 cars car NOUN fcis-6350 52 33 . . PUNCT fcis-6350 53 1 the the DET fcis-6350 53 2 common common ADJ fcis-6350 53 3 metric metric NOUN fcis-6350 53 4 for for ADP fcis-6350 53 5 detection detection NOUN fcis-6350 53 6 speed speed NOUN fcis-6350 53 7 is be AUX fcis-6350 53 8 frame frame NOUN fcis-6350 53 9 per per ADP fcis-6350 53 10 second second ADJ fcis-6350 53 11 ( ( PUNCT fcis-6350 53 12 fps fps PROPN fcis-6350 53 13 ) ) PUNCT fcis-6350 53 14 , , PUNCT fcis-6350 53 15 which which PRON fcis-6350 53 16 indicates indicate VERB fcis-6350 53 17 the the DET fcis-6350 53 18 number number NOUN fcis-6350 53 19 of of ADP fcis-6350 53 20 images image NOUN fcis-6350 53 21 that that PRON fcis-6350 53 22 can can AUX fcis-6350 53 23 be be AUX fcis-6350 53 24 detected detect VERB fcis-6350 53 25 per per ADP fcis-6350 53 26 second second NOUN fcis-6350 53 27 , , PUNCT fcis-6350 53 28 and and CCONJ fcis-6350 53 29 the the DET fcis-6350 53 30 larger large ADJ fcis-6350 53 31 the the DET fcis-6350 53 32 fps fps PROPN fcis-6350 53 33 , , PUNCT fcis-6350 53 34 the the PRON fcis-6350 53 35 faster fast ADV fcis-6350 53 36 the the DET fcis-6350 53 37 model model NOUN fcis-6350 53 38 detects detect NOUN fcis-6350 53 39 and and CCONJ fcis-6350 53 40 the the PRON fcis-6350 53 41 better well ADJ fcis-6350 53 42 the the DET fcis-6350 53 43 real real ADJ fcis-6350 53 44 - - PUNCT fcis-6350 53 45 time time NOUN fcis-6350 53 46 performance performance NOUN fcis-6350 53 47 . . PUNCT fcis-6350 54 1 3.6 3.6 NUM fcis-6350 54 2 . . PUNCT fcis-6350 54 3 training training NOUN fcis-6350 54 4 strategies strategy NOUN fcis-6350 54 5 this this DET fcis-6350 54 6 experiment experiment NOUN fcis-6350 54 7 was be AUX fcis-6350 54 8 conducted conduct VERB fcis-6350 54 9 on on ADP fcis-6350 54 10 the the DET fcis-6350 54 11 dataset dataset NOUN fcis-6350 54 12 kitti kitti PROPN fcis-6350 54 13 , , PUNCT fcis-6350 54 14 and and CCONJ fcis-6350 54 15 5000 5000 NUM fcis-6350 54 16 images image NOUN fcis-6350 54 17 in in ADP fcis-6350 54 18 kitti kitti PROPN fcis-6350 54 19 were be AUX fcis-6350 54 20 selected select VERB fcis-6350 54 21 and and CCONJ fcis-6350 54 22 divided divide VERB fcis-6350 54 23 into into ADP fcis-6350 54 24 training training NOUN fcis-6350 54 25 set set NOUN fcis-6350 54 26 , , PUNCT fcis-6350 54 27 validation validation NOUN fcis-6350 54 28 set set NOUN fcis-6350 54 29 , , PUNCT fcis-6350 54 30 and and CCONJ fcis-6350 54 31 test test NOUN fcis-6350 54 32 set set VERB fcis-6350 54 33 in in ADP fcis-6350 54 34 the the DET fcis-6350 54 35 ratio ratio NOUN fcis-6350 54 36 of of ADP fcis-6350 54 37 8:1:1 8:1:1 PROPN fcis-6350 54 38 . . PUNCT fcis-6350 55 1 the the DET fcis-6350 55 2 divided divided ADJ fcis-6350 55 3 training training NOUN fcis-6350 55 4 set set NOUN fcis-6350 55 5 is be AUX fcis-6350 55 6 4000 4000 NUM fcis-6350 55 7 images image NOUN fcis-6350 55 8 , , PUNCT fcis-6350 55 9 and and CCONJ fcis-6350 55 10 the the DET fcis-6350 55 11 validation validation NOUN fcis-6350 55 12 set set NOUN fcis-6350 55 13 and and CCONJ fcis-6350 55 14 test test NOUN fcis-6350 55 15 set set VERB fcis-6350 55 16 are be AUX fcis-6350 55 17 500 500 NUM fcis-6350 55 18 images image NOUN fcis-6350 55 19 each each PRON fcis-6350 55 20 . . PUNCT fcis-6350 56 1 the the DET fcis-6350 56 2 input input NOUN fcis-6350 56 3 image image NOUN fcis-6350 56 4 size size NOUN fcis-6350 56 5 is be AUX fcis-6350 56 6 640ร—640 640ร—640 NUM fcis-6350 56 7 , , PUNCT fcis-6350 56 8 the the DET fcis-6350 56 9 batchsize batchsize NOUN fcis-6350 56 10 is be AUX fcis-6350 56 11 16 16 NUM fcis-6350 56 12 , , PUNCT fcis-6350 56 13 the the DET fcis-6350 56 14 test test NOUN fcis-6350 56 15 batchsize batchsize NOUN fcis-6350 56 16 is be AUX fcis-6350 56 17 1 1 NUM fcis-6350 56 18 , , PUNCT fcis-6350 56 19 the the DET fcis-6350 56 20 initial initial ADJ fcis-6350 56 21 learning learning NOUN fcis-6350 56 22 rate rate NOUN fcis-6350 56 23 is be AUX fcis-6350 56 24 0.01 0.01 NUM fcis-6350 56 25 , , PUNCT fcis-6350 56 26 the the DET fcis-6350 56 27 learning learning NOUN fcis-6350 56 28 rate rate NOUN fcis-6350 56 29 momentum momentum NOUN fcis-6350 56 30 is be AUX fcis-6350 56 31 0.937 0.937 NUM fcis-6350 56 32 , , PUNCT fcis-6350 56 33 and and CCONJ fcis-6350 56 34 the the DET fcis-6350 56 35 training training NOUN fcis-6350 56 36 epoch epoch NOUN fcis-6350 56 37 is be AUX fcis-6350 56 38 200 200 NUM fcis-6350 56 39 rounds round NOUN fcis-6350 56 40 . . PUNCT fcis-6350 57 1 adaptive adaptive ADJ fcis-6350 57 2 anchor anchor NOUN fcis-6350 57 3 is be AUX fcis-6350 57 4 used use VERB fcis-6350 57 5 to to PART fcis-6350 57 6 set set VERB fcis-6350 57 7 the the DET fcis-6350 57 8 initial initial ADJ fcis-6350 57 9 anchor anchor NOUN fcis-6350 57 10 box box NOUN fcis-6350 57 11 size size NOUN fcis-6350 57 12 , , PUNCT fcis-6350 57 13 adaptive adaptive ADJ fcis-6350 57 14 moment moment NOUN fcis-6350 57 15 estimation estimation NOUN fcis-6350 57 16 ( ( PUNCT fcis-6350 57 17 adam adam PROPN fcis-6350 57 18 ) ) PUNCT fcis-6350 57 19 optimizer optimizer NOUN fcis-6350 57 20 is be AUX fcis-6350 57 21 used use VERB fcis-6350 57 22 to to PART fcis-6350 57 23 optimize optimize VERB fcis-6350 57 24 the the DET fcis-6350 57 25 model model NOUN fcis-6350 57 26 , , PUNCT fcis-6350 57 27 and and CCONJ fcis-6350 57 28 mosaic mosaic ADJ fcis-6350 57 29 data datum NOUN fcis-6350 57 30 augmentation augmentation NOUN fcis-6350 57 31 is be AUX fcis-6350 57 32 used use VERB fcis-6350 57 33 to to PART fcis-6350 57 34 augment augment VERB fcis-6350 57 35 the the DET fcis-6350 57 36 dataset dataset NOUN fcis-6350 57 37 , , PUNCT fcis-6350 57 38 which which PRON fcis-6350 57 39 increases increase VERB fcis-6350 57 40 the the DET fcis-6350 57 41 diversity diversity NOUN fcis-6350 57 42 of of ADP fcis-6350 57 43 the the DET fcis-6350 57 44 dataset dataset NOUN fcis-6350 57 45 and and CCONJ fcis-6350 57 46 the the DET fcis-6350 57 47 number number NOUN fcis-6350 57 48 of of ADP fcis-6350 57 49 targets target NOUN fcis-6350 57 50 , , PUNCT fcis-6350 57 51 so so SCONJ fcis-6350 57 52 that that SCONJ fcis-6350 57 53 the the DET fcis-6350 57 54 algorithm algorithm NOUN fcis-6350 57 55 has have VERB fcis-6350 57 56 better well ADJ fcis-6350 57 57 generalization generalization NOUN fcis-6350 57 58 ability ability NOUN fcis-6350 57 59 . . PUNCT fcis-6350 58 1 the the DET fcis-6350 58 2 other other ADJ fcis-6350 58 3 parameters parameter NOUN fcis-6350 58 4 are be AUX fcis-6350 58 5 set set VERB fcis-6350 58 6 according accord VERB fcis-6350 58 7 to to ADP fcis-6350 58 8 the the DET fcis-6350 58 9 official official ADJ fcis-6350 58 10 default default NOUN fcis-6350 58 11 parameters parameter NOUN fcis-6350 58 12 of of ADP fcis-6350 58 13 yolov5s yolov5s PROPN fcis-6350 58 14 . . PUNCT fcis-6350 59 1 4 4 X fcis-6350 59 2 . . NUM fcis-6350 59 3 experiments experiment NOUN fcis-6350 59 4 4.1 4.1 NUM fcis-6350 59 5 . . PUNCT fcis-6350 60 1 lightweight lightweight ADJ fcis-6350 60 2 improvement improvement NOUN fcis-6350 60 3 experiment experiment NOUN fcis-6350 60 4 in in ADP fcis-6350 60 5 order order NOUN fcis-6350 60 6 to to PART fcis-6350 60 7 verify verify VERB fcis-6350 60 8 the the DET fcis-6350 60 9 effectiveness effectiveness NOUN fcis-6350 60 10 of of ADP fcis-6350 60 11 lightweight lightweight ADJ fcis-6350 60 12 networks network NOUN fcis-6350 61 1 , , PUNCT fcis-6350 61 2 the the DET fcis-6350 61 3 efficientnetv2 efficientnetv2 PROPN fcis-6350 61 4 lightweight lightweight ADJ fcis-6350 61 5 network network NOUN fcis-6350 61 6 was be AUX fcis-6350 61 7 replaced replace VERB fcis-6350 61 8 with with ADP fcis-6350 61 9 the the DET fcis-6350 61 10 original original ADJ fcis-6350 61 11 yolov5s yolov5s PROPN fcis-6350 61 12 and and CCONJ fcis-6350 61 13 yolov5 yolov5 NOUN fcis-6350 61 14 - - PUNCT fcis-6350 61 15 nano nano PROPN fcis-6350 61 16 and and CCONJ fcis-6350 61 17 mobilenetv3 mobilenetv3 NOUN fcis-6350 61 18 were be AUX fcis-6350 61 19 replaced replace VERB fcis-6350 61 20 with with ADP fcis-6350 61 21 the the DET fcis-6350 61 22 original original ADJ fcis-6350 61 23 yolov5s yolov5s PROPN fcis-6350 61 24 network network NOUN fcis-6350 61 25 . . PUNCT fcis-6350 62 1 comparative comparative ADJ fcis-6350 62 2 experiments experiment NOUN fcis-6350 62 3 were be AUX fcis-6350 62 4 conducted conduct VERB fcis-6350 62 5 using use VERB fcis-6350 62 6 the the DET fcis-6350 62 7 same same ADJ fcis-6350 62 8 dataset dataset NOUN fcis-6350 62 9 under under ADP fcis-6350 62 10 the the DET fcis-6350 62 11 same same ADJ fcis-6350 62 12 configuration configuration NOUN fcis-6350 62 13 environment environment NOUN fcis-6350 62 14 . . PUNCT fcis-6350 63 1 the the DET fcis-6350 63 2 comparison comparison NOUN fcis-6350 63 3 results result NOUN fcis-6350 63 4 are be AUX fcis-6350 63 5 shown show VERB fcis-6350 63 6 in in ADP fcis-6350 63 7 table table NOUN fcis-6350 63 8 2 2 NUM fcis-6350 63 9 table table NOUN fcis-6350 63 10 2 2 NUM fcis-6350 63 11 . . PUNCT fcis-6350 63 12 lightweight lightweight ADJ fcis-6350 63 13 experimental experimental ADJ fcis-6350 63 14 results result NOUN fcis-6350 63 15 models model NOUN fcis-6350 63 16 backbone backbone NOUN fcis-6350 63 17 size size NOUN fcis-6350 63 18 par par PROPN fcis-6350 63 19 map0.5 map0.5 PROPN fcis-6350 63 20 fps fps PROPN fcis-6350 63 21 yolov5s yolov5s PROPN fcis-6350 63 22 csp1_x csp1_x VERB fcis-6350 63 23 14.4 14.4 NUM fcis-6350 63 24 7.07 7.07 NUM fcis-6350 63 25 83.5 83.5 NUM fcis-6350 63 26 60 60 NUM fcis-6350 63 27 ynano ynano NOUN fcis-6350 63 28 csp1_x csp1_x VERB fcis-6350 63 29 3.7 3.7 NUM fcis-6350 63 30 1.76 1.76 NUM fcis-6350 63 31 78.2 78.2 NUM fcis-6350 63 32 93 93 NUM fcis-6350 63 33 ymbv3 ymbv3 NOUN fcis-6350 64 1 mobv3 mobv3 NOUN fcis-6350 64 2 7.3 7.3 NUM fcis-6350 64 3 3.55 3.55 NUM fcis-6350 64 4 76.1 76.1 NUM fcis-6350 64 5 73 73 NUM fcis-6350 64 6 yeffv2 yeffv2 NOUN fcis-6350 64 7 effv2 effv2 NOUN fcis-6350 64 8 11.2 11.2 NUM fcis-6350 64 9 5.43 5.43 NUM fcis-6350 64 10 81.4 81.4 NUM fcis-6350 64 11 66 66 NUM fcis-6350 64 12 yolov5s yolov5s NOUN fcis-6350 64 13 - - PUNCT fcis-6350 64 14 nano nano PROPN fcis-6350 64 15 is be AUX fcis-6350 64 16 short short ADJ fcis-6350 64 17 as as ADP fcis-6350 64 18 ynano ynano PROPN fcis-6350 64 19 , , PUNCT fcis-6350 64 20 yolov5s yolov5s PROPN fcis-6350 64 21 - - PUNCT fcis-6350 64 22 mobilenetv3 mobilenetv3 PROPN fcis-6350 64 23 is be AUX fcis-6350 64 24 short short ADJ fcis-6350 64 25 as as SCONJ fcis-6350 64 26 ymbv3 ymbv3 PROPN fcis-6350 64 27 yolov5s yolov5s PROPN fcis-6350 64 28 - - PUNCT fcis-6350 64 29 efficientnetv2 efficientnetv2 PROPN fcis-6350 64 30 is be AUX fcis-6350 64 31 short short ADJ fcis-6350 64 32 as as ADP fcis-6350 64 33 yeffv2 yeffv2 NOUN fcis-6350 64 34 . . PUNCT fcis-6350 65 1 from from ADP fcis-6350 65 2 the the DET fcis-6350 65 3 results result NOUN fcis-6350 65 4 of of ADP fcis-6350 65 5 the the DET fcis-6350 65 6 lightweight lightweight ADJ fcis-6350 65 7 experiments experiment NOUN fcis-6350 65 8 in in ADP fcis-6350 65 9 table table NOUN fcis-6350 65 10 2 2 NUM fcis-6350 65 11 , , PUNCT fcis-6350 65 12 we we PRON fcis-6350 65 13 can can AUX fcis-6350 65 14 see see VERB fcis-6350 65 15 that that SCONJ fcis-6350 65 16 the the DET fcis-6350 65 17 performance performance NOUN fcis-6350 65 18 of of ADP fcis-6350 65 19 all all DET fcis-6350 65 20 networks network NOUN fcis-6350 65 21 tested test VERB fcis-6350 65 22 in in ADP fcis-6350 65 23 the the DET fcis-6350 65 24 data datum NOUN fcis-6350 65 25 set set VERB fcis-6350 65 26 , , PUNCT fcis-6350 65 27 yolov5s yolov5s PROPN fcis-6350 65 28 has have VERB fcis-6350 65 29 the the DET fcis-6350 65 30 highest high ADJ fcis-6350 65 31 indexes index NOUN fcis-6350 65 32 , , PUNCT fcis-6350 65 33 with with ADP fcis-6350 65 34 map@0.5 map@0.5 NOUN fcis-6350 65 35 reaching reach VERB fcis-6350 65 36 83.5 83.5 NUM fcis-6350 65 37 ; ; PUNCT fcis-6350 65 38 among among ADP fcis-6350 65 39 the the DET fcis-6350 65 40 lightweight lightweight ADJ fcis-6350 65 41 networks network NOUN fcis-6350 65 42 , , PUNCT fcis-6350 65 43 yolov5smobilenetv3 yolov5smobilenetv3 PROPN fcis-6350 65 44 has have VERB fcis-6350 65 45 the the DET fcis-6350 65 46 lowest low ADJ fcis-6350 65 47 accuracy accuracy NOUN fcis-6350 65 48 rate rate NOUN fcis-6350 65 49 , , PUNCT fcis-6350 65 50 map@0.5 map@0.5 NOUN fcis-6350 65 51 value value NOUN fcis-6350 65 52 among among ADP fcis-6350 65 53 all all DET fcis-6350 65 54 networks network NOUN fcis-6350 65 55 , , PUNCT fcis-6350 65 56 and and CCONJ fcis-6350 65 57 the the DET fcis-6350 65 58 map@0.5 map@0.5 PROPN fcis-6350 65 59 of of ADP fcis-6350 65 60 yolov5sefficientnetv2 yolov5sefficientnetv2 PROPN fcis-6350 65 61 reaches reach VERB fcis-6350 65 62 81.4 81.4 NUM fcis-6350 65 63 , , PUNCT fcis-6350 65 64 which which PRON fcis-6350 65 65 is be AUX fcis-6350 65 66 higher high ADJ fcis-6350 65 67 than than ADP fcis-6350 65 68 the the DET fcis-6350 65 69 76.1 76.1 NUM fcis-6350 65 70 of of ADP fcis-6350 65 71 yolov5s yolov5s PROPN fcis-6350 65 72 - - PUNCT fcis-6350 65 73 mobilenetv3 mobilenetv3 PROPN fcis-6350 65 74 network network NOUN fcis-6350 65 75 and and CCONJ fcis-6350 65 76 78.2 78.2 NUM fcis-6350 65 77 of of ADP fcis-6350 65 78 yolov5s yolov5s PROPN fcis-6350 65 79 - - PUNCT fcis-6350 65 80 nano nano NOUN fcis-6350 65 81 , , PUNCT fcis-6350 65 82 and and CCONJ fcis-6350 65 83 is be AUX fcis-6350 65 84 the the DET fcis-6350 65 85 best good ADJ fcis-6350 65 86 performance performance NOUN fcis-6350 65 87 among among ADP fcis-6350 65 88 all all DET fcis-6350 65 89 lightweight lightweight ADJ fcis-6350 65 90 networks network NOUN fcis-6350 65 91 . . PUNCT fcis-6350 66 1 22.2 22.2 NUM fcis-6350 66 2 % % NOUN fcis-6350 66 3 , , PUNCT fcis-6350 66 4 23.2 23.2 NUM fcis-6350 66 5 % % NOUN fcis-6350 66 6 , , PUNCT fcis-6350 66 7 and and CCONJ fcis-6350 66 8 the the DET fcis-6350 66 9 fps fps NOUN fcis-6350 66 10 is be AUX fcis-6350 66 11 higher high ADJ fcis-6350 66 12 than than ADP fcis-6350 66 13 the the DET fcis-6350 66 14 60 60 NUM fcis-6350 66 15 frames frame NOUN fcis-6350 66 16 per per ADP fcis-6350 66 17 second second NOUN fcis-6350 66 18 of of ADP fcis-6350 66 19 the the DET fcis-6350 66 20 original original ADJ fcis-6350 66 21 yolov5s yolov5s PROPN fcis-6350 66 22 . . PUNCT fcis-6350 67 1 it it PRON fcis-6350 67 2 is be AUX fcis-6350 67 3 slightly slightly ADV fcis-6350 67 4 lower low ADJ fcis-6350 67 5 than than ADP fcis-6350 67 6 the the DET fcis-6350 67 7 73 73 NUM fcis-6350 67 8 frames frame NOUN fcis-6350 67 9 per per ADP fcis-6350 67 10 second second NOUN fcis-6350 67 11 of of ADP fcis-6350 67 12 yolov5s yolov5s PROPN fcis-6350 67 13 - - PUNCT fcis-6350 67 14 mobilenetv3 mobilenetv3 PROPN fcis-6350 67 15 . . PUNCT fcis-6350 68 1 4.2 4.2 NUM fcis-6350 68 2 . . PUNCT fcis-6350 68 3 add add VERB fcis-6350 68 4 odconv odconv ADJ fcis-6350 68 5 moudle moudle NOUN fcis-6350 68 6 in in ADP fcis-6350 68 7 order order NOUN fcis-6350 68 8 to to PART fcis-6350 68 9 compensate compensate VERB fcis-6350 68 10 for for ADP fcis-6350 68 11 the the DET fcis-6350 68 12 loss loss NOUN fcis-6350 68 13 of of ADP fcis-6350 68 14 accuracy accuracy NOUN fcis-6350 68 15 after after ADP fcis-6350 68 16 light light ADJ fcis-6350 68 17 weighting weighting NOUN fcis-6350 68 18 , , PUNCT fcis-6350 68 19 the the DET fcis-6350 68 20 od od PROPN fcis-6350 68 21 dynamic dynamic ADJ fcis-6350 68 22 convolution convolution NOUN fcis-6350 68 23 is be AUX fcis-6350 68 24 fused fuse VERB fcis-6350 68 25 in in ADP fcis-6350 68 26 the the DET fcis-6350 68 27 light light NOUN fcis-6350 68 28 weighting weighting NOUN fcis-6350 68 29 network network NOUN fcis-6350 68 30 and and CCONJ fcis-6350 68 31 the the DET fcis-6350 68 32 comparison comparison NOUN fcis-6350 68 33 experiments experiment NOUN fcis-6350 68 34 are be AUX fcis-6350 68 35 conducted conduct VERB fcis-6350 68 36 as as SCONJ fcis-6350 68 37 follows follow VERB fcis-6350 68 38 . . PUNCT fcis-6350 69 1 yolov5s yolov5s PROPN fcis-6350 69 2 - - PUNCT fcis-6350 69 3 efficientnetv2 efficientnetv2 NOUN fcis-6350 69 4 - - PUNCT fcis-6350 69 5 od od PROPN fcis-6350 69 6 is be AUX fcis-6350 69 7 abbreviated abbreviate VERB fcis-6350 69 8 as as ADP fcis-6350 69 9 yeffv2 yeffv2 NOUN fcis-6350 69 10 - - PUNCT fcis-6350 69 11 od od NOUN fcis-6350 69 12 . . PUNCT fcis-6350 70 1 from from ADP fcis-6350 70 2 table table NOUN fcis-6350 70 3 3 3 NUM fcis-6350 70 4 , , PUNCT fcis-6350 70 5 we we PRON fcis-6350 70 6 can can AUX fcis-6350 70 7 see see VERB fcis-6350 70 8 that that SCONJ fcis-6350 70 9 after after ADP fcis-6350 70 10 adding add VERB fcis-6350 70 11 odconv odconv ADJ fcis-6350 70 12 convolution convolution NOUN fcis-6350 70 13 to to ADP fcis-6350 70 14 the the DET fcis-6350 70 15 lightweight lightweight ADJ fcis-6350 70 16 network network NOUN fcis-6350 70 17 yolov5s yolov5s PROPN fcis-6350 70 18 efficientnetv2 efficientnetv2 PROPN fcis-6350 70 19 , , PUNCT fcis-6350 70 20 the the DET fcis-6350 70 21 map@0.5 map@0.5 NOUN fcis-6350 70 22 value value NOUN fcis-6350 70 23 improved improve VERB fcis-6350 70 24 by by ADP fcis-6350 70 25 1.4 1.4 NUM fcis-6350 70 26 % % NOUN fcis-6350 70 27 with with ADP fcis-6350 70 28 137 137 NUM fcis-6350 70 29 better well ADJ fcis-6350 70 30 detection detection NOUN fcis-6350 70 31 capability capability NOUN fcis-6350 70 32 for for ADP fcis-6350 70 33 small small ADJ fcis-6350 70 34 targets target NOUN fcis-6350 70 35 , , PUNCT fcis-6350 70 36 although although SCONJ fcis-6350 70 37 the the DET fcis-6350 70 38 fps fps NOUN fcis-6350 70 39 decreased decrease VERB fcis-6350 70 40 slightly slightly ADV fcis-6350 70 41 by by ADP fcis-6350 70 42 7 7 NUM fcis-6350 70 43 frames frame NOUN fcis-6350 70 44 per per ADP fcis-6350 70 45 second second NOUN fcis-6350 70 46 . . PUNCT fcis-6350 71 1 the the DET fcis-6350 71 2 original original ADJ fcis-6350 71 3 image image NOUN fcis-6350 71 4 and and CCONJ fcis-6350 71 5 the the DET fcis-6350 71 6 detection detection NOUN fcis-6350 71 7 effect effect NOUN fcis-6350 71 8 are be AUX fcis-6350 71 9 shown show VERB fcis-6350 71 10 in in ADP fcis-6350 71 11 figure figure NOUN fcis-6350 71 12 8 8 NUM fcis-6350 71 13 and and CCONJ fcis-6350 71 14 figure figure VERB fcis-6350 71 15 9 9 NUM fcis-6350 71 16 table3 table3 NOUN fcis-6350 71 17 . . PUNCT fcis-6350 72 1 add add VERB fcis-6350 72 2 odconv odconv ADJ fcis-6350 72 3 experimental experimental ADJ fcis-6350 72 4 results result NOUN fcis-6350 72 5 models model NOUN fcis-6350 72 6 odconv odconv VERB fcis-6350 72 7 map0.5 map0.5 PROPN fcis-6350 72 8 fps fps PROPN fcis-6350 72 9 yolov5s yolov5s PROPN fcis-6350 72 10 83.5 83.5 NUM fcis-6350 72 11 60 60 NUM fcis-6350 72 12 ynano ynano NOUN fcis-6350 72 13 78.2 78.2 NUM fcis-6350 72 14 93 93 NUM fcis-6350 72 15 ymbv3 ymbv3 NOUN fcis-6350 73 1 76.1 76.1 NUM fcis-6350 73 2 73 73 NUM fcis-6350 73 3 yeffv2 yeffv2 NOUN fcis-6350 73 4 81.4 81.4 NUM fcis-6350 73 5 66 66 NUM fcis-6350 73 6 yeffv2 yeffv2 NOUN fcis-6350 73 7 - - PUNCT fcis-6350 73 8 od od NUM fcis-6350 73 9 โˆš โˆš PROPN fcis-6350 73 10 82.8 82.8 NUM fcis-6350 73 11 59 59 NUM fcis-6350 73 12 figure figure NOUN fcis-6350 73 13 8 8 NUM fcis-6350 73 14 . . PUNCT fcis-6350 74 1 original original ADJ fcis-6350 74 2 image image NOUN fcis-6350 74 3 figure figure NOUN fcis-6350 74 4 9 9 NUM fcis-6350 74 5 . . PUNCT fcis-6350 75 1 detection detection NOUN fcis-6350 75 2 effect effect NOUN fcis-6350 75 3 picture picture NOUN fcis-6350 75 4 5 5 NUM fcis-6350 75 5 . . PUNCT fcis-6350 75 6 conclusion conclusion NOUN fcis-6350 75 7 in in ADP fcis-6350 75 8 this this DET fcis-6350 75 9 paper paper NOUN fcis-6350 75 10 , , PUNCT fcis-6350 75 11 firstly firstly ADV fcis-6350 75 12 , , PUNCT fcis-6350 75 13 the the DET fcis-6350 75 14 efficientnetv2 efficientnetv2 PROPN fcis-6350 75 15 lightweight lightweight ADJ fcis-6350 75 16 network network NOUN fcis-6350 75 17 replaces replace VERB fcis-6350 75 18 the the DET fcis-6350 75 19 yolov5s yolov5s PROPN fcis-6350 75 20 backbone backbone NOUN fcis-6350 75 21 network network NOUN fcis-6350 75 22 , , PUNCT fcis-6350 75 23 which which PRON fcis-6350 75 24 reduces reduce VERB fcis-6350 75 25 the the DET fcis-6350 75 26 number number NOUN fcis-6350 75 27 of of ADP fcis-6350 75 28 network network NOUN fcis-6350 75 29 model model NOUN fcis-6350 75 30 parameters parameter NOUN fcis-6350 75 31 and and CCONJ fcis-6350 75 32 computation computation NOUN fcis-6350 75 33 , , PUNCT fcis-6350 75 34 and and CCONJ fcis-6350 75 35 secondly secondly ADV fcis-6350 75 36 , , PUNCT fcis-6350 75 37 the the DET fcis-6350 75 38 odconv odconv ADJ fcis-6350 75 39 dynamic dynamic ADJ fcis-6350 75 40 convolution convolution NOUN fcis-6350 75 41 is be AUX fcis-6350 75 42 added add VERB fcis-6350 75 43 to to PART fcis-6350 75 44 make make VERB fcis-6350 75 45 the the DET fcis-6350 75 46 model model NOUN fcis-6350 75 47 feature feature NOUN fcis-6350 75 48 extraction extraction NOUN fcis-6350 75 49 ability ability NOUN fcis-6350 75 50 stronger strong ADJ fcis-6350 75 51 , , PUNCT fcis-6350 75 52 after after ADP fcis-6350 75 53 experimental experimental ADJ fcis-6350 75 54 comparison comparison NOUN fcis-6350 75 55 , , PUNCT fcis-6350 75 56 the the DET fcis-6350 75 57 improved improved ADJ fcis-6350 75 58 yolov5s yolov5s PROPN fcis-6350 75 59 efficientnetv2 efficientnetv2 PROPN fcis-6350 75 60 - - PUNCT fcis-6350 75 61 od od PROPN fcis-6350 75 62 network network NOUN fcis-6350 75 63 model model NOUN fcis-6350 75 64 map@0.5 map@0.5 PROPN fcis-6350 75 65 reaches reach VERB fcis-6350 75 66 82.8 82.8 NUM fcis-6350 75 67 and and CCONJ fcis-6350 75 68 the the DET fcis-6350 75 69 pfs pfs PROPN fcis-6350 75 70 is be AUX fcis-6350 75 71 59 59 NUM fcis-6350 75 72 , , PUNCT fcis-6350 75 73 which which PRON fcis-6350 75 74 reduces reduce VERB fcis-6350 75 75 the the DET fcis-6350 75 76 requirements requirement NOUN fcis-6350 75 77 for for ADP fcis-6350 75 78 the the DET fcis-6350 75 79 performance performance NOUN fcis-6350 75 80 of of ADP fcis-6350 75 81 on on ADP fcis-6350 75 82 - - PUNCT fcis-6350 75 83 board board NOUN fcis-6350 75 84 devices device NOUN fcis-6350 75 85 while while SCONJ fcis-6350 75 86 ensuring ensure VERB fcis-6350 75 87 the the DET fcis-6350 75 88 lightweight lightweight NOUN fcis-6350 75 89 and and CCONJ fcis-6350 75 90 detection detection NOUN fcis-6350 75 91 accuracy accuracy NOUN fcis-6350 75 92 . . PUNCT fcis-6350 76 1 references reference NOUN fcis-6350 76 2 [ [ X fcis-6350 76 3 1 1 X fcis-6350 76 4 ] ] X fcis-6350 76 5 girshick girshick ADJ fcis-6350 76 6 r r PROPN fcis-6350 76 7 , , PUNCT fcis-6350 76 8 donahue donahue PROPN fcis-6350 76 9 j j PROPN fcis-6350 76 10 , , PUNCT fcis-6350 76 11 darrell darrell PROPN fcis-6350 76 12 t t PROPN fcis-6350 76 13 , , PUNCT fcis-6350 76 14 et et PROPN fcis-6350 76 15 al al PROPN fcis-6350 76 16 . . PROPN fcis-6350 77 1 rich rich ADJ fcis-6350 77 2 feature feature NOUN fcis-6350 77 3 hierarchies hierarchy NOUN fcis-6350 77 4 for for ADP fcis-6350 77 5 accurate accurate ADJ fcis-6350 77 6 object object NOUN fcis-6350 77 7 detection detection NOUN fcis-6350 77 8 and and CCONJ fcis-6350 77 9 semantic semantic ADJ fcis-6350 77 10 segmentation segmentation NOUN fcis-6350 78 1 [ [ X fcis-6350 78 2 j j X fcis-6350 78 3 ] ] X fcis-6350 78 4 . . PUNCT fcis-6350 79 1 2013 2013 NUM fcis-6350 79 2 . . PUNCT fcis-6350 80 1 [ [ X fcis-6350 80 2 2 2 NUM fcis-6350 80 3 ] ] X fcis-6350 80 4 girshick girshick PROPN fcis-6350 80 5 r. r. PROPN fcis-6350 80 6 2015 2015 NUM fcis-6350 80 7 ieee ieee PROPN fcis-6350 80 8 international international ADJ fcis-6350 80 9 conference conference NOUN fcis-6350 80 10 on on ADP fcis-6350 80 11 computer computer NOUN fcis-6350 80 12 vision vision NOUN fcis-6350 80 13 ( ( PUNCT fcis-6350 80 14 iccv iccv PROPN fcis-6350 80 15 ) ) PUNCT fcis-6350 80 16 ; ; PUNCT fcis-6350 80 17 proceedings proceeding NOUN fcis-6350 80 18 of of ADP fcis-6350 80 19 the the DET fcis-6350 80 20 ieee ieee NOUN fcis-6350 80 21 international international PROPN fcis-6350 80 22 conference conference NOUN fcis-6350 80 23 on on ADP fcis-6350 80 24 computer computer NOUN fcis-6350 80 25 vision vision NOUN fcis-6350 80 26 , , PUNCT fcis-6350 80 27 f f PROPN fcis-6350 80 28 , , PUNCT fcis-6350 80 29 2015 2015 NUM fcis-6350 81 1 [ [ X fcis-6350 81 2 c c X fcis-6350 81 3 ] ] PUNCT fcis-6350 81 4 . . PUNCT fcis-6350 82 1 [ [ X fcis-6350 82 2 3 3 X fcis-6350 82 3 ] ] X fcis-6350 82 4 ren ren PROPN fcis-6350 82 5 s s PROPN fcis-6350 82 6 , , PUNCT fcis-6350 82 7 he he PRON fcis-6350 82 8 k k NOUN fcis-6350 82 9 , , PUNCT fcis-6350 82 10 girshick girshick ADJ fcis-6350 82 11 r r NOUN fcis-6350 82 12 , , PUNCT fcis-6350 82 13 et et PROPN fcis-6350 82 14 al al PROPN fcis-6350 82 15 . . PUNCT fcis-6350 83 1 faster fast ADJ fcis-6350 83 2 r r NOUN fcis-6350 83 3 - - PUNCT fcis-6350 83 4 cnn cnn NOUN fcis-6350 83 5 : : PUNCT fcis-6350 83 6 towards towards ADP fcis-6350 83 7 real real ADJ fcis-6350 83 8 - - PUNCT fcis-6350 83 9 time time NOUN fcis-6350 83 10 object object NOUN fcis-6350 83 11 detection detection NOUN fcis-6350 83 12 with with ADP fcis-6350 83 13 region region NOUN fcis-6350 83 14 proposal proposal NOUN fcis-6350 83 15 networks network NOUN fcis-6350 83 16 ; ; PUNCT fcis-6350 83 17 proceedings proceeding NOUN fcis-6350 83 18 of of ADP fcis-6350 83 19 the the DET fcis-6350 83 20 nips nip NOUN fcis-6350 83 21 , , PUNCT fcis-6350 83 22 f f PROPN fcis-6350 83 23 , , PUNCT fcis-6350 83 24 2016 2016 NUM fcis-6350 84 1 [ [ X fcis-6350 84 2 c c X fcis-6350 84 3 ] ] PUNCT fcis-6350 84 4 . . PUNCT fcis-6350 85 1 [ [ X fcis-6350 85 2 4 4 NUM fcis-6350 85 3 ] ] X fcis-6350 85 4 wei wei PROPN fcis-6350 85 5 l l PROPN fcis-6350 85 6 , , PUNCT fcis-6350 85 7 dragomir dragomir VERB fcis-6350 85 8 a a PRON fcis-6350 85 9 , , PUNCT fcis-6350 85 10 dumitru dumitru PROPN fcis-6350 85 11 e e NOUN fcis-6350 85 12 , , PUNCT fcis-6350 85 13 et et PROPN fcis-6350 85 14 al al PROPN fcis-6350 85 15 . . PROPN fcis-6350 85 16 ssd ssd PROPN fcis-6350 85 17 : : PUNCT fcis-6350 85 18 single single ADJ fcis-6350 85 19 shot shot NOUN fcis-6350 85 20 multibox multibox NOUN fcis-6350 85 21 detector detector NOUN fcis-6350 86 1 [ [ X fcis-6350 86 2 j j X fcis-6350 86 3 ] ] X fcis-6350 86 4 . . PUNCT fcis-6350 87 1 2016 2016 NUM fcis-6350 87 2 . . PUNCT fcis-6350 88 1 [ [ X fcis-6350 88 2 5 5 X fcis-6350 88 3 ] ] X fcis-6350 88 4 redmon redmon PROPN fcis-6350 88 5 j j PROPN fcis-6350 88 6 , , PUNCT fcis-6350 88 7 divvala divvala PROPN fcis-6350 88 8 s s PROPN fcis-6350 88 9 , , PUNCT fcis-6350 88 10 girshick girshick ADJ fcis-6350 88 11 r r NOUN fcis-6350 88 12 , , PUNCT fcis-6350 88 13 et et PROPN fcis-6350 88 14 al al PROPN fcis-6350 88 15 . . PUNCT fcis-6350 89 1 you you PRON fcis-6350 89 2 only only ADV fcis-6350 89 3 look look VERB fcis-6350 89 4 once once ADV fcis-6350 89 5 : : PUNCT fcis-6350 89 6 unified unified ADJ fcis-6350 89 7 , , PUNCT fcis-6350 89 8 real real ADJ fcis-6350 89 9 - - PUNCT fcis-6350 89 10 time time NOUN fcis-6350 89 11 object object NOUN fcis-6350 89 12 detection detection NOUN fcis-6350 89 13 ; ; PUNCT fcis-6350 89 14 proceedings proceeding NOUN fcis-6350 89 15 of of ADP fcis-6350 89 16 the the DET fcis-6350 89 17 computer computer NOUN fcis-6350 89 18 vision vision NOUN fcis-6350 89 19 & & CCONJ fcis-6350 89 20 pattern pattern NOUN fcis-6350 89 21 recognition recognition NOUN fcis-6350 89 22 , , PUNCT fcis-6350 89 23 f f X fcis-6350 89 24 , , PUNCT fcis-6350 89 25 2016 2016 NUM fcis-6350 90 1 [ [ X fcis-6350 90 2 c c X fcis-6350 90 3 ] ] PUNCT fcis-6350 90 4 . . PUNCT fcis-6350 91 1 [ [ X fcis-6350 91 2 6 6 NUM fcis-6350 91 3 ] ] X fcis-6350 91 4 redmon redmon PROPN fcis-6350 91 5 j j PROPN fcis-6350 91 6 , , PUNCT fcis-6350 91 7 farhadi farhadi PROPN fcis-6350 91 8 a. a. PROPN fcis-6350 91 9 yolo9000 yolo9000 PROPN fcis-6350 91 10 : : PUNCT fcis-6350 91 11 better well ADJ fcis-6350 91 12 , , PUNCT fcis-6350 91 13 faster fast ADJ fcis-6350 91 14 , , PUNCT fcis-6350 91 15 stronger strong ADJ fcis-6350 91 16 ; ; PUNCT fcis-6350 91 17 proceedings proceeding NOUN fcis-6350 91 18 of of ADP fcis-6350 91 19 the the DET fcis-6350 91 20 ieee ieee NOUN fcis-6350 91 21 conference conference NOUN fcis-6350 91 22 on on ADP fcis-6350 91 23 computer computer NOUN fcis-6350 91 24 vision vision NOUN fcis-6350 91 25 & & CCONJ fcis-6350 91 26 pattern pattern NOUN fcis-6350 91 27 recognition recognition NOUN fcis-6350 91 28 , , PUNCT fcis-6350 91 29 f f PROPN fcis-6350 91 30 , , PUNCT fcis-6350 91 31 2017 2017 NUM fcis-6350 92 1 [ [ X fcis-6350 92 2 c c X fcis-6350 92 3 ] ] PUNCT fcis-6350 92 4 . . PUNCT fcis-6350 93 1 [ [ X fcis-6350 93 2 7 7 X fcis-6350 93 3 ] ] X fcis-6350 93 4 redmon redmon PROPN fcis-6350 93 5 j j PROPN fcis-6350 93 6 , , PUNCT fcis-6350 93 7 farhadi farhadi NOUN fcis-6350 93 8 a a DET fcis-6350 93 9 j j PROPN fcis-6350 93 10 a a DET fcis-6350 93 11 e e NOUN fcis-6350 93 12 - - NOUN fcis-6350 93 13 p. p. NOUN fcis-6350 93 14 yolov3 yolov3 PROPN fcis-6350 93 15 : : PUNCT fcis-6350 93 16 an an DET fcis-6350 93 17 incremental incremental ADJ fcis-6350 93 18 improvement improvement NOUN fcis-6350 94 1 [ [ X fcis-6350 94 2 j j X fcis-6350 94 3 ] ] X fcis-6350 94 4 . . PUNCT fcis-6350 95 1 2018 2018 NUM fcis-6350 95 2 . . PUNCT fcis-6350 96 1 [ [ X fcis-6350 96 2 8 8 NUM fcis-6350 96 3 ] ] X fcis-6350 96 4 bochkovskiy bochkovskiy X fcis-6350 96 5 a a PRON fcis-6350 96 6 , , PUNCT fcis-6350 96 7 wang wang PROPN fcis-6350 96 8 c c PROPN fcis-6350 96 9 y y PROPN fcis-6350 96 10 , , PUNCT fcis-6350 96 11 liao liao PROPN fcis-6350 96 12 h. h. PROPN fcis-6350 96 13 yolov4 yolov4 PROPN fcis-6350 97 1 : : PUNCT fcis-6350 97 2 optimal optimal ADJ fcis-6350 97 3 speed speed NOUN fcis-6350 97 4 and and CCONJ fcis-6350 97 5 accuracy accuracy NOUN fcis-6350 97 6 of of ADP fcis-6350 97 7 object object NOUN fcis-6350 97 8 detection detection NOUN fcis-6350 98 1 [ [ X fcis-6350 98 2 j j X fcis-6350 98 3 ] ] X fcis-6350 98 4 . . PUNCT fcis-6350 99 1 2020 2020 NUM fcis-6350 99 2 . . PUNCT fcis-6350 100 1 [ [ X fcis-6350 100 2 9 9 NUM fcis-6350 100 3 ] ] X fcis-6350 100 4 ge ge PROPN fcis-6350 100 5 z z PROPN fcis-6350 100 6 , , PUNCT fcis-6350 100 7 liu liu PROPN fcis-6350 100 8 s s PROPN fcis-6350 100 9 , , PUNCT fcis-6350 100 10 wang wang PROPN fcis-6350 100 11 f f PROPN fcis-6350 100 12 , , PUNCT fcis-6350 100 13 et et PROPN fcis-6350 100 14 al al PROPN fcis-6350 100 15 . . PROPN fcis-6350 100 16 yolox yolox PROPN fcis-6350 100 17 : : PUNCT fcis-6350 100 18 exceeding exceed VERB fcis-6350 100 19 yolo yolo ADJ fcis-6350 100 20 series series NOUN fcis-6350 100 21 in in ADP fcis-6350 100 22 2021 2021 NUM fcis-6350 101 1 [ [ X fcis-6350 101 2 j j X fcis-6350 101 3 ] ] X fcis-6350 101 4 . . PUNCT fcis-6350 102 1 2021 2021 NUM fcis-6350 102 2 . . PUNCT fcis-6350 103 1 [ [ X fcis-6350 103 2 10 10 NUM fcis-6350 103 3 ] ] X fcis-6350 103 4 ren ren PROPN fcis-6350 103 5 j j PROPN fcis-6350 103 6 c c NOUN fcis-6350 103 7 x x SYM fcis-6350 103 8 h h PROPN fcis-6350 103 9 , , PUNCT fcis-6350 103 10 liu liu PROPN fcis-6350 103 11 j j PROPN fcis-6350 103 12 b b PROPN fcis-6350 103 13 , , PUNCT fcis-6350 103 14 et et PROPN fcis-6350 103 15 al al PROPN fcis-6350 103 16 . . PROPN fcis-6350 103 17 accurate accurate ADJ fcis-6350 103 18 single single ADJ fcis-6350 103 19 stage stage NOUN fcis-6350 103 20 detector detector NOUN fcis-6350 103 21 using use VERB fcis-6350 103 22 recurrent recurrent ADJ fcis-6350 103 23 rolling rolling ADJ fcis-6350 103 24 convolution convolution NOUN fcis-6350 103 25 [ [ X fcis-6350 103 26 j j X fcis-6350 103 27 ] ] X fcis-6350 103 28 . . PUNCT fcis-6350 104 1 ieee ieee PROPN fcis-6350 104 2 conference conference PROPN fcis-6350 104 3 on on ADP fcis-6350 104 4 computer computer NOUN fcis-6350 104 5 vision vision NOUN fcis-6350 104 6 and and CCONJ fcis-6350 104 7 pattern pattern NOUN fcis-6350 104 8 recognition recognition NOUN fcis-6350 104 9 , , PUNCT fcis-6350 104 10 2017 2017 NUM fcis-6350 104 11 , , PUNCT fcis-6350 104 12 honolulu honolulu PROPN fcis-6350 104 13 , , PUNCT fcis-6350 104 14 hi hi PROPN fcis-6350 104 15 , , PUNCT fcis-6350 104 16 usa usa PROPN fcis-6350 104 17 . . PROPN fcis-6350 104 18 new new PROPN fcis-6350 104 19 york york PROPN fcis-6350 104 20 : : PUNCT fcis-6350 104 21 ieee ieee PROPN fcis-6350 104 22 press press PROPN fcis-6350 104 23 , , PUNCT fcis-6350 104 24 : : PUNCT fcis-6350 104 25 752 752 NUM fcis-6350 104 26 - - SYM fcis-6350 104 27 60 60 NUM fcis-6350 104 28 . . PUNCT fcis-6350 105 1 [ [ X fcis-6350 105 2 11 11 NUM fcis-6350 105 3 ] ] X fcis-6350 105 4 al al PROPN fcis-6350 105 5 l l NOUN fcis-6350 105 6 x x PUNCT fcis-6350 105 7 w w PROPN fcis-6350 105 8 w w PROPN fcis-6350 105 9 w w PROPN fcis-6350 105 10 l l PROPN fcis-6350 105 11 e. e. PROPN fcis-6350 105 12 generalized generalize VERB fcis-6350 105 13 focal focal ADJ fcis-6350 105 14 loss loss NOUN fcis-6350 105 15 ๏ผš ๏ผš PUNCT fcis-6350 105 16 learning learn VERB fcis-6350 105 17 qualified qualified ADJ fcis-6350 105 18 and and CCONJ fcis-6350 105 19 distributed distribute VERB fcis-6350 105 20 bounding bounding NOUN fcis-6350 105 21 boxes box NOUN fcis-6350 105 22 for for ADP fcis-6350 105 23 dense dense ADJ fcis-6350 105 24 object object NOUN fcis-6350 105 25 detection detection NOUN fcis-6350 105 26 . . PUNCT fcis-6350 106 1 advances advance NOUN fcis-6350 106 2 in in ADP fcis-6350 106 3 neural neural ADJ fcis-6350 106 4 information information NOUN fcis-6350 106 5 processing processing NOUN fcis-6350 106 6 systems system NOUN fcis-6350 106 7 , , PUNCT fcis-6350 106 8 2020 2020 NUM fcis-6350 106 9 . . PUNCT fcis-6350 107 1 [ [ X fcis-6350 107 2 12 12 NUM fcis-6350 107 3 ] ] X fcis-6350 107 4 huang huang PROPN fcis-6350 107 5 wenhan wenhan PROPN fcis-6350 107 6 y y PROPN fcis-6350 107 7 g g PROPN fcis-6350 107 8 , , PUNCT fcis-6350 107 9 geng geng PROPN fcis-6350 107 10 keke keke PROPN fcis-6350 107 11 , , PUNCT fcis-6350 107 12 et et PROPN fcis-6350 107 13 al al PROPN fcis-6350 107 14 . . PROPN fcis-6350 107 15 target target NOUN fcis-6350 107 16 detection detection NOUN fcis-6350 107 17 in in ADP fcis-6350 107 18 complex complex ADJ fcis-6350 107 19 driving driving NOUN fcis-6350 107 20 scenes scene NOUN fcis-6350 107 21 based base VERB fcis-6350 107 22 on on ADP fcis-6350 107 23 adaptive adaptive ADJ fcis-6350 107 24 fusion fusion NOUN fcis-6350 107 25 of of ADP fcis-6350 107 26 dilated dilated ADJ fcis-6350 107 27 convolutional convolutional ADJ fcis-6350 107 28 features feature NOUN fcis-6350 107 29 [ [ X fcis-6350 107 30 j j X fcis-6350 107 31 ] ] X fcis-6350 107 32 . . PUNCT fcis-6350 108 1 journal journal PROPN fcis-6350 108 2 of of ADP fcis-6350 108 3 southeast southeast PROPN fcis-6350 108 4 university university PROPN fcis-6350 108 5 natural natural ADJ fcis-6350 108 6 science science NOUN fcis-6350 108 7 edition edition NOUN fcis-6350 108 8 , , PUNCT fcis-6350 108 9 2021 2021 NUM fcis-6350 108 10 , , PUNCT fcis-6350 108 11 ( ( PUNCT fcis-6350 108 12 51(06 51(06 NOUN fcis-6350 108 13 ) ) PUNCT fcis-6350 108 14 ): ): PUNCT fcis-6350 108 15 1076 1076 NUM fcis-6350 108 16 - - SYM fcis-6350 108 17 83 83 NUM fcis-6350 108 18 . . PUNCT fcis-6350 109 1 [ [ X fcis-6350 109 2 13 13 NUM fcis-6350 109 3 ] ] X fcis-6350 110 1 li li PROPN fcis-6350 110 2 chao chao PROPN fcis-6350 110 3 z z PROPN fcis-6350 110 4 a a DET fcis-6350 110 5 y y PROPN fcis-6350 110 6 a. a. NOUN fcis-6350 110 7 omni omni ADJ fcis-6350 110 8 - - PUNCT fcis-6350 110 9 dimensional dimensional ADJ fcis-6350 110 10 dynamic dynamic ADJ fcis-6350 110 11 convolution convolution NOUN fcis-6350 110 12 [ [ X fcis-6350 110 13 eb eb PROPN fcis-6350 110 14 / / SYM fcis-6350 110 15 ol ol PROPN fcis-6350 110 16 ] ] X fcis-6350 110 17 https://arxivorg https://arxivorg PROPN fcis-6350 110 18 / / SYM fcis-6350 110 19 pdf/220907947pdf pdf/220907947pdf NOUN fcis-6350 110 20 , , PUNCT fcis-6350 110 21 [ [ X fcis-6350 110 22 2022 2022 NUM fcis-6350 110 23 - - SYM fcis-6350 110 24 09 09 NUM fcis-6350 110 25 - - SYM fcis-6350 110 26 11 11 NUM fcis-6350 110 27 ] ] PUNCT fcis-6350 110 28 . . PUNCT fcis-6350 111 1 [ [ X fcis-6350 111 2 14 14 NUM fcis-6350 111 3 ] ] X fcis-6350 111 4 lin lin PROPN fcis-6350 111 5 t t PROPN fcis-6350 111 6 y y PROPN fcis-6350 111 7 , , PUNCT fcis-6350 111 8 dollรกr dollรกr NOUN fcis-6350 111 9 p p NOUN fcis-6350 111 10 , , PUNCT fcis-6350 111 11 girshick girshick ADJ fcis-6350 111 12 r r NOUN fcis-6350 111 13 , , PUNCT fcis-6350 111 14 et et PROPN fcis-6350 111 15 al al PROPN fcis-6350 111 16 . . PROPN fcis-6350 112 1 feature feature PROPN fcis-6350 112 2 pyramid pyramid NOUN fcis-6350 112 3 networks network NOUN fcis-6350 112 4 for for ADP fcis-6350 112 5 object object NOUN fcis-6350 112 6 detection detection NOUN fcis-6350 113 1 [ [ X fcis-6350 113 2 j j X fcis-6350 113 3 ] ] X fcis-6350 113 4 . . PUNCT fcis-6350 114 1 2016 2016 NUM fcis-6350 114 2 . . PUNCT fcis-6350 115 1 [ [ X fcis-6350 115 2 15 15 NUM fcis-6350 115 3 ] ] X fcis-6350 115 4 liu liu PROPN fcis-6350 115 5 s s PROPN fcis-6350 115 6 , , PUNCT fcis-6350 115 7 qi qi PROPN fcis-6350 115 8 l l NOUN fcis-6350 115 9 , , PUNCT fcis-6350 115 10 qin qin PROPN fcis-6350 115 11 h h PROPN fcis-6350 115 12 , , PUNCT fcis-6350 115 13 et et PROPN fcis-6350 115 14 al al PROPN fcis-6350 115 15 . . PROPN fcis-6350 115 16 path path PROPN fcis-6350 115 17 aggregation aggregation NOUN fcis-6350 115 18 network network NOUN fcis-6350 115 19 for for ADP fcis-6350 115 20 instance instance NOUN fcis-6350 115 21 segmentation segmentation NOUN fcis-6350 116 1 [ [ X fcis-6350 116 2 j j X fcis-6350 116 3 ] ] X fcis-6350 116 4 . . PUNCT fcis-6350 117 1 2018 2018 NUM fcis-6350 117 2 . . PUNCT fcis-6350 118 1 [ [ X fcis-6350 118 2 16 16 NUM fcis-6350 118 3 ] ] X fcis-6350 118 4 q q PROPN fcis-6350 119 1 t t PROPN fcis-6350 119 2 m m PROPN fcis-6350 119 3 l. l. PROPN fcis-6350 119 4 efficientnetv2 efficientnetv2 PROPN fcis-6350 119 5 : : PUNCT fcis-6350 119 6 smaller small ADJ fcis-6350 119 7 models model NOUN fcis-6350 119 8 and and CCONJ fcis-6350 119 9 faster fast ADJ fcis-6350 119 10 training training NOUN fcis-6350 120 1 [ [ X fcis-6350 120 2 j j X fcis-6350 120 3 ] ] X fcis-6350 120 4 . . PUNCT fcis-6350 121 1 international international ADJ fcis-6350 121 2 conference conference NOUN fcis-6350 121 3 on on ADP fcis-6350 121 4 machine machine NOUN fcis-6350 121 5 learning learning NOUN fcis-6350 121 6 , , PUNCT fcis-6350 121 7 2021 2021 NUM fcis-6350 121 8 : : PUNCT fcis-6350 121 9 10096106 10096106 NUM fcis-6350 121 10 . . PUNCT fcis-6350 122 1 [ [ X fcis-6350 122 2 17 17 NUM fcis-6350 122 3 ] ] X fcis-6350 122 4 c c PROPN fcis-6350 122 5 t t PROPN fcis-6350 122 6 m m VERB fcis-6350 122 7 l l NOUN fcis-6350 122 8 q q PUNCT fcis-6350 123 1 e e NOUN fcis-6350 123 2 r r NOUN fcis-6350 123 3 m m NOUN fcis-6350 123 4 s s NOUN fcis-6350 123 5 f f X fcis-6350 123 6 c c PROPN fcis-6350 123 7 n n PROPN fcis-6350 123 8 n. n. PROPN fcis-6350 123 9 efficientnet efficientnet PROPN fcis-6350 123 10 : : PUNCT fcis-6350 123 11 rethinking rethink VERB fcis-6350 123 12 model model NOUN fcis-6350 123 13 scaling scale VERB fcis-6350 123 14 for for ADP fcis-6350 123 15 convolutional convolutional ADJ fcis-6350 123 16 neural neural ADJ fcis-6350 123 17 networks network NOUN fcis-6350 123 18 [ [ X fcis-6350 123 19 j j X fcis-6350 123 20 ] ] X fcis-6350 123 21 . . PUNCT fcis-6350 124 1 proceedings proceeding NOUN fcis-6350 124 2 of of ADP fcis-6350 124 3 the the DET fcis-6350 124 4 36th 36th ADJ fcis-6350 124 5 international international ADJ fcis-6350 124 6 conference conference NOUN fcis-6350 124 7 on on ADP fcis-6350 124 8 machine machine NOUN fcis-6350 124 9 learning learning NOUN fcis-6350 124 10 , , PUNCT fcis-6350 124 11 2019 2019 NUM fcis-6350 124 12 : : PUNCT fcis-6350 124 13 6105 6105 NUM fcis-6350 124 14 - - PUNCT fcis-6350 124 15 14.r 14.r NUM fcis-6350 124 16 https://arxivorg https://arxivorg PROPN fcis-6350 124 17 / / SYM fcis-6350 124 18 pdf/220907947pdf pdf/220907947pdf NOUN