Transcription of Large Scale Visual Recognition Challenge (ILSVRC) 2017
1 Large Scale Visual Recognition Challenge ( ilsvrc ) 2017 Eunbyung ParkUNC Chapel HillOverviewWei LiuUNC Chapel HillOlga RussakovskyCMU/PrincetonJiaDengUniv. of MichiganFei-FeiLiStanfordAlex BergUNC Chapel over the +CLS Task Task Task ResultsParticipation in ilsvrc over the years35152981123157172115201020112012201 32014201520162017 The number of Entries1 year9 monthILSVRC Image Classification (CLS) TaskSteel drum1000 object classes1,431,167 imagesCLS-LOCILSVRC Image Classification (CLS) TaskSteel drumILSVRC Image Localization (LOC) TaskSteel drumILSVRC Image Localization (LOC) TaskSteel drumCorrectBad localizationBad classificationILSVRC Image Localization (LOC) TaskSteel drumCorrectClassification Results (CLS) Localization Results (LOC)Localization Team NameError(%) , Li Shen2, Gang Sun11. Momenta2. Universifyof OxfordTrimps-SouchenXiaotengZhang, ZhengyanDing, JianyingZhou, JieShao, Lin Mei The Third Research Institute of the Ministry of Public Security, CLS Results - Provided DataTeam NameError(%) Shuai1, ZhenboYu1, QingshanLiu1, XiaotongYuan1, KaihuaZhang1, YishengZhu1, GuangcanLiu1, Jing Yang1, YuxiangZhou2, JiankangDeng21.
2 Nanjing University of Information Science & Technology2. Imperial College LondonNUS-Qihoo_DPNsYunpengChen1, HuaxinXiao1, JiananLi1, XuechengNie1, XiaojieJin1, JianshuLi1, JiashiFeng1, Jian Dong2, ShuichengYan21. NUS -National University of Singapore2. Qihoo360 ILSVRC2017 CLS Results - External DataILSVRC2017 LOC Results - Provided DataTeam NameError(%) , HuaxinXiao1, JiananLi1, XuechengNie1, XiaojieJin1, JianshuLi1, JiashiFeng1, Jian Dong2, ShuichengYan21. NUS -National University of Singapore2. Qihoo360 Trimps-SouchenXiaotengZhang, ZhengyanDing, JianyingZhou, JieShao, Lin Mei The Third Research Institute of the Ministry of Public Security, NameError(%) Shuai1, ZhenboYu1, QingshanLiu1, XiaotongYuan1, KaihuaZhang1, YishengZhu1, GuangcanLiu1, Jing Yang1, YuxiangZhou2, JiankangDeng21. Nanjing University of Information Science & Technology2. Imperial College LondonNUS-Qihoo_DPNsYunpengChen1, HuaxinXiao1, JiananLi1, XuechengNie1, XiaojieJin1, JianshuLi1, JiashiFeng1, Jian Dong2, ShuichengYan21.
3 NUS -National University of Singapore2. Qihoo360 ILSVRC2017 LOC Results - External DataILSVRC Object Detection (DET) Task200 object classes578,482 imagesDETILSVRC Object Detection (DET) TaskThis year: 5,500 new test images with bounding boxes fully annotatedBoxes are correct if IoU> Average PrecisionIoU=RecallPrecisionArea under Precision Recall Curves011 Detection Results (DET)Mean Average Precision(mAP) DET Results - Provided DataTeam Name#category wonmAP(%) Shuai1, ZhenboYu1, QingshanLiu1, XiaotongYuan1, KaihuaZhang1, YishengZhu1, GuangcanLiu1, Jing Yang1, YuxiangZhou2, JiankangDeng21. Nanjing University of Information Science & Technology2. Imperial College LondonNUS-Qihoo_DPNsYunpengChen1, HuaxinXiao1, JiananLi1, XuechengNie1, XiaojieJin1, JianshuLi1, JiashiFeng1, Jian Dong2, ShuichengYan21. NUS -National University of Singapore2. Qihoo360 ILSVRC2017 DET Results - External DataTeam Name#category wonmAP(%) Shuai1, ZhenboYu1, QingshanLiu1, XiaotongYuan1, KaihuaZhang1, YishengZhu1, GuangcanLiu1, Jing Yang1, YuxiangZhou2, JiankangDeng21.
4 Nanjing University of Information Science & Technology2. Imperial College LondonNUS-Qihoo_DPNsYunpengChen1, HuaxinXiao1, JiananLi1, XuechengNie1, XiaojieJin1, JianshuLi1, JiashiFeng1, Jian Dong2, ShuichengYan21. NUS -National University of Singapore2. Qihoo360 Object Detection from Video(VID) TaskAllows evaluation of generic object detectionin cluttered videos at scaleFully annotated 30 object classes across 7,314 snippetsObject Detection from Video(VID) TaskThis year: 1,036 new snippets distributed into train, val, test set. Algorithms outputs a list of bounding box detections with confidences A detection is considered correct if intersection over union(IoU) overlap with ground truth > Evaluated by average precision per object class Winner of Challenge is the team that wins the most object categoriesEvaluation modeled after PASCAL VOC: Object Detection from Video(VID) TaskThis year: 1,036 new snippets distributed into train, val, test set.
5 Algorithms outputs a list of bounding box detections with confidences and trackletID. Trackletsare sorted by the mean confidence. A trackletis considered correct if intersection over union(IoU) overlap with ground truth tracklet> Evaluation by average precision per class. Final score is an average over different thresholds. Winner of Challenge is the team that has highest taking tracking into account: Video Detection Results (VID)Mean Average Precision(mAP) TrackingILSVRC2017 VID Results - Provided DataTeam Name#category wonmAP(%)mAP(%)trackingIC& (VID) , YuxiangZhou1, BaoshengYu2, ZheChen2, StefanosZafeiriou1, DachengTao2, 1. Imperial College London2. University of SydneyNUS-Qihoo-UIUC_DPNs(VID)YunchaoWei 1, MengdanZhang1, JiananLi1, YunpengChen1, JiashiFeng1, Jian Dong2, ShuichengYan2, HonghuiShi31. National University of Singapore2. Qihoo3603.
6 University of Illinois Urbana-ChampaignILSVRC2017 VID Results - External DataTeam Name#category wonmAP(%)mAP(%)trackingIC& (VID) , YuxiangZhou1, BaoshengYu2, ZheChen2, StefanosZafeiriou1, DachengTao21. Imperial College London2. University of SydneyNUS-Qihoo-UIUC_DPNs(VID)YunchaoWei 1, MengdanZhang1, JiananLi1, YunpengChen1, JiashiFeng1, Jian Dong2, ShuichengYan2, HonghuiShi31. National University of Singapore2. Qihoo3603. University of Illinois Urbana-ChampaignComing Presentations! (Team: WMW, Momenta): Squeeze-and-Excitation (Team: NUS-Qihoo_DPNs, NUS): Dual Path Networks and its presentations of winning entries: NUS-Qihoo-UIUC_DPNs (VID), DeepView(ETRI), MIL_UT, SIIT_KAIST-SKT, KAISTNIA_ETRI
