DejaVid

Reported on 8 benchmarks across 3 tasks

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

Robots3 results

Activity RecognitiononHMDB-51
Average accuracy of 3 splits· uses extra data
88.6
best: 88.7 (VideoMAE V2-g)
Activity RecognitiononSomething-Something V2
Top-1 Accuracy· uses extra data
77.2
best: 77.3 (MVD (Kinetics400 pretrain, ViT-H, 16 frame))
Activity RecognitiononSomething-Something V2
Top-5 Accuracy· uses extra data
96.3

Time Series3 results

Action RecognitiononHMDB-51
Average accuracy of 3 splits· uses extra data
88.6
best: 88.7 (VideoMAE V2-g)
Action RecognitiononSomething-Something V2
Top-1 Accuracy· uses extra data
77.2
best: 77.3 (MVD (Kinetics400 pretrain, ViT-H, 16 frame))
Action RecognitiononSomething-Something V2
Top-5 Accuracy· uses extra data
96.3

Computer Vision2 results

VideoonKinetics-400
Acc@1· uses extra data
89.1
best: 93.6 (OmniVec2)
VideoonKinetics-400
Acc@5· uses extra data
98.2
best: 98.9 (TubeViT-H (ImageNet-1k))