MARS+RGB+Flow (64 frames, Kinetics pretrained)

Reported on 4 benchmarks across 2 tasks

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

Robots2 results

Activity RecognitiononSomething-Something V1
Top 1 Accuracy· uses extra data
53
best: 70 (InternVideo)
Activity RecognitiononUCF101
3-fold Accuracy· uses extra data
97.8
best: 99.7 (FTP-UniFormerV2-L/14)

Time Series2 results

Action RecognitiononSomething-Something V1
Top 1 Accuracy· uses extra data
53
best: 70 (InternVideo)
Action RecognitiononUCF101
3-fold Accuracy· uses extra data
97.8
best: 99.7 (FTP-UniFormerV2-L/14)