Two-Stream I3D (Imagenet+Kinetics pre-training)

Reported on 4 benchmarks across 2 tasks · 1 paper · 2 SOTA

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

Robots2 results

Activity RecognitiononUCF101
3-fold Accuracy· 2017-05-22
98
best: 99.7 (FTP-UniFormerV2-L/14)
SOTA
Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset arXiv:1705.07750
Activity RecognitiononHMDB-51
Average accuracy of 3 splits· uses extra data· 2017-05-22
80.7
best: 88.7 (VideoMAE V2-g)
Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset arXiv:1705.07750

Time Series2 results

Action RecognitiononUCF101
3-fold Accuracy· 2017-05-22
98
best: 99.7 (FTP-UniFormerV2-L/14)
SOTA
Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset arXiv:1705.07750
Action RecognitiononHMDB-51
Average accuracy of 3 splits· uses extra data· 2017-05-22
80.7
best: 88.7 (VideoMAE V2-g)
Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset arXiv:1705.07750