Faster-RCNN + two-stream I3D conv

Reported on 4 benchmarks across 1 task · 1 paper · 4 SOTA

Note: results are matched by exact model name. Different papers may use the same name for different model variants.

Time Series4 results

Action DetectiononUCF101-24
Frame-mAP 0.5· 2017-05-23
76.3
best: 90.3 (STAR/L)
SOTA
AVA: A Video Dataset of Spatio-temporally Localized Atomic Visual Actions arXiv:1705.08421
Action DetectiononUCF101-24
Video-mAP 0.5· 2017-05-23
59.9
best: 76.3 (Stable Mean Teacher (I3D))
SOTA
AVA: A Video Dataset of Spatio-temporally Localized Atomic Visual Actions arXiv:1705.08421
Action DetectiononJ-HMDB
Frame-mAP 0.5· 2017-05-23
73.3
best: 88.5 (SiA)
SOTA
AVA: A Video Dataset of Spatio-temporally Localized Atomic Visual Actions arXiv:1705.08421
Action DetectiononJ-HMDB
Video-mAP 0.5· 2017-05-23
78.6
best: 88.1 (HIT)
SOTA
AVA: A Video Dataset of Spatio-temporally Localized Atomic Visual Actions arXiv:1705.08421