Action Recognition on EPIC-KITCHENS VISOR
42.82Pred. Acc. VerbDense
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| DenseBackbone=ActionCLIP ViT-B/16, P-Ratio (%)=0, GFLOPs=24.1592026.06 | 42.82 | 32.21 | 33.91 | 38.13 | 12.99 | 98.44 | 45.59 | 37.71 | |
| DenseModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=0, GFLOPs=7.5122026.06 | 38.1 | 28.23 | — | — | — | — | — | — | |
| Rationale-Informed PruningBackbone=ActionCLIP ViT-B/16, P-Ratio (%)=20, GFLOPs=18.5112026.06 | 36.87 | 26.68 | 37.58 | 31.75 | 11.93 | 99.37 | 36.56 | 31.54 | |
| Rationale-Informed PruningModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=20, GFLOPs=6.5932026.06 | 35.93 | 27.52 | — | — | — | — | — | — | |
| DenseBackbone=CLIP ViT-B/16, P-Ratio (%)=0, GFLOPs=24.1592026.06 | 34.38 | 31.82 | 2.93 | 32.69 | 1.13 | 100 | 42.01 | 38.5 | |
| Rationale-Informed PruningBackbone=ActionCLIP ViT-B/16, P-Ratio (%)=25, GFLOPs=17.6502026.06 | 33.76 | 24.09 | 37.7 | 28.11 | 11.07 | 99.32 | 34.06 | 29.16 | |
| Rationale-Informed PruningBackbone=ActionCLIP ViT-B/16, P-Ratio (%)=30, GFLOPs=16.6612026.06 | 31.89 | 21.47 | 35.79 | 24.19 | 9.34 | 99.49 | 30.25 | 25.98 | |
| Rationale-Informed PruningModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=25, GFLOPs=5.8982026.06 | 31.86 | 22.28 | — | — | — | — | — | — | |
| DenseBackbone=CLIP ViT-B/32, P-Ratio (%)=0, GFLOPs=7.5122026.06 | 30.76 | 26.65 | 10.8 | 40.3 | 3.58 | 98.8 | 37.4 | 32.74 | |
| Rationale-Informed PruningBackbone=CLIP ViT-B/16, P-Ratio (%)=20, GFLOPs=18.4202026.06 | 29.24 | 24.26 | 8.17 | 35.24 | 2.45 | 99.56 | 34.56 | 29.89 | |
| Rationale-Informed PruningModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=30, GFLOPs=5.6302026.06 | 28.65 | 18.95 | — | — | — | — | — | — | |
| Rationale-Informed PruningBackbone=CLIP ViT-B/32, P-Ratio (%)=20, GFLOPs=6.1172026.06 | 26.78 | 21.96 | 10.23 | 33.33 | 2.92 | 99.07 | 33.15 | 28.23 | |
| Rationale-Informed PruningBackbone=CLIP ViT-B/16, P-Ratio (%)=25, GFLOPs=17.5532026.06 | 26.16 | 22.67 | 9.26 | 35.09 | 2.65 | 99.07 | 33.52 | 28.36 | |
| OMPBackbone=CLIP ViT-B/16, P-Ratio (%)=25, GFLOPs=18.1172026.06 | 25.19 | 0.84 | 0 | 0 | 0 | 75 | 0.58 | 0.2 | |
| Rationale-Informed PruningBackbone=CLIP ViT-B/32, P-Ratio (%)=25, GFLOPs=5.8792026.06 | 25.03 | 19.4 | 11.02 | 28.97 | 2.78 | 99.74 | 29.34 | 25.18 | |
| ECoFLaPBackbone=ActionCLIP ViT-B/16, P-Ratio (%)=20, GFLOPs=17.7132026.06 | 24.32 | 14.78 | 39.86 | 17.78 | 7.42 | 98.22 | 27.68 | 18.29 | |
| ECoFLaPBackbone=CLIP ViT-B/16, P-Ratio (%)=20, GFLOPs=16.8082026.06 | 23.93 | 25.32 | 6.84 | 38.64 | 2.25 | 99.8 | 36.88 | 32.87 | |
| ECoFLaPModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=25, GFLOPs=5.7652026.06 | 23.8 | 17.95 | — | — | — | — | — | — | |
| Kwon et al.Backbone=ActionCLIP ViT-B/16, P-Ratio (%)=20, GFLOPs=19.7002026.06 | 23.64 | 14.1 | 29.86 | 18.53 | 5.5 | 97.12 | 24.66 | 17.89 | |
| ECoFLaPModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=20, GFLOPs=5.7602026.06 | 23.22 | 17.21 | — | — | — | — | — | — | |
| Rationale-Informed PruningBackbone=CLIP ViT-B/16, P-Ratio (%)=30, GFLOPs=16.5802026.06 | 23.16 | 20.67 | 11 | 27.92 | 2.85 | 99.23 | 31.21 | 25.71 | |
| Rationale-Informed PruningBackbone=CLIP ViT-B/32, P-Ratio (%)=30, GFLOPs=5.6202026.06 | 20.7 | 15.56 | 13.87 | 24.43 | 2.85 | 100 | 23.92 | 20.54 | |
| ECoFLaPBackbone=CLIP ViT-B/16, P-Ratio (%)=25, GFLOPs=15.8722026.06 | 20.7 | 22.25 | 5.97 | 33.78 | 1.66 | 99.76 | 32.13 | 27.7 | |
| GMPBackbone=CLIP ViT-B/16, P-Ratio (%)=25, GFLOPs=21.2152026.06 | 19.4 | 2.1 | 3.7 | 2.27 | 0.07 | 88.89 | 12.83 | 1.59 | |
| ECoFLaPModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=30, GFLOPs=5.4382026.06 | 18.6 | 10.25 | — | — | — | — | — | — | |
| GMPBackbone=CLIP ViT-B/16, P-Ratio (%)=30, GFLOPs=20.7082026.06 | 17.79 | 1.97 | 0 | 0 | 0 | 96.15 | 18.38 | 1.66 | |
| GMPBackbone=CLIP ViT-B/16, P-Ratio (%)=20, GFLOPs=21.7022026.06 | 17.69 | 2.72 | 3.57 | 0.79 | 0.07 | 100 | 9.49 | 1.86 | |
| OMPBackbone=CLIP ViT-B/16, P-Ratio (%)=20, GFLOPs=19.3262026.06 | 16.88 | 0.81 | 0 | 0 | 0 | 100 | 0.51 | 0.27 | |
| GMPBackbone=CLIP ViT-B/32, P-Ratio (%)=30, GFLOPs=6.6312026.06 | 16.24 | 1.2 | 4.35 | 0.78 | 0.07 | 65.22 | 9.26 | 0.99 | |
| ECoFLaPBackbone=CLIP ViT-B/32, P-Ratio (%)=20, GFLOPs=5.9012026.06 | 16.04 | 14 | 8.7 | 16.3 | 1.46 | 99.21 | 18.94 | 16.63 | |
| GMPBackbone=CLIP ViT-B/32, P-Ratio (%)=20, GFLOPs=6.8782026.06 | 14.59 | 1.78 | 0 | 0 | 0 | 100 | 4.75 | 1.06 | |
| ECoFLaPBackbone=CLIP ViT-B/32, P-Ratio (%)=25, GFLOPs=5.7762026.06 | 14.46 | 13.32 | 12.2 | 23.85 | 2.05 | 99.61 | 19.28 | 16.77 | |
| Kwon et al.Model=ActionCLIP ViT-B/32, Pruning Ratio (%)=20, GFLOPs=6.3802026.06 | 13.75 | 9.41 | — | — | — | — | — | — | |
| Kwon et al.Backbone=CLIP ViT-B/32, P-Ratio (%)=20, GFLOPs=6.3802026.06 | 13.58 | 9.38 | 9.55 | 9.71 | 1.13 | 99.44 | 16.15 | 11.73 | |
| Kwon et al.Backbone=CLIP ViT-B/32, P-Ratio (%)=30, GFLOPs=5.8132026.06 | 13.29 | 2.65 | 6.06 | 1.23 | 0.13 | 93.94 | 4.61 | 2.05 | |
| ECoFLaPBackbone=CLIP ViT-B/16, P-Ratio (%)=30, GFLOPs=15.6052026.06 | 13.23 | 16.27 | 4.84 | 25.86 | 0.99 | 99.03 | 25.91 | 20.34 | |
| Kwon et al.Backbone=CLIP ViT-B/16, P-Ratio (%)=20, GFLOPs=19.7002026.06 | 13.13 | 13.97 | 2.13 | 17.65 | 0.4 | 97.87 | 21.87 | 18.29 | |
| ECoFLaPBackbone=ActionCLIP ViT-B/16, P-Ratio (%)=30, GFLOPs=15.0012026.06 | 11.71 | 4.46 | 32.88 | 5.02 | 1.59 | 94.52 | 9.69 | 4.57 | |
| Kwon et al.Backbone=CLIP ViT-B/32, P-Ratio (%)=25, GFLOPs=6.0962026.06 | 11.19 | 4.46 | 4.11 | 1.55 | 0.2 | 97.26 | 7.68 | 4.71 | |
| ECoFLaPBackbone=CLIP ViT-B/32, P-Ratio (%)=30, GFLOPs=5.5042026.06 | 10.64 | 9.83 | 10.5 | 9.69 | 1.26 | 98.9 | 14.88 | 11.86 | |
| OMPBackbone=CLIP ViT-B/16, P-Ratio (%)=30, GFLOPs=16.9092026.06 | 8.67 | 0.94 | 0 | 0 | 0 | 85.71 | 1.22 | 0.4 | |
| ECoFLaPBackbone=ActionCLIP ViT-B/16, P-Ratio (%)=25, GFLOPs=15.4762026.06 | 8.51 | 4.72 | 36.9 | 6.1 | 2.05 | 92.86 | 12.4 | 5.17 | |
| Kwon et al.Backbone=ActionCLIP ViT-B/16, P-Ratio (%)=25, GFLOPs=18.5822026.06 | 8.25 | 4.2 | 41.33 | 6.65 | 2.05 | 97.33 | 8.83 | 4.84 | |
| GMPModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=20, GFLOPs=6.8792026.06 | 7.76 | 0.52 | — | — | — | — | — | — | |
| Kwon et al.Model=ActionCLIP ViT-B/32, Pruning Ratio (%)=25, GFLOPs=6.0962026.06 | 7.24 | 4.24 | — | — | — | — | — | — | |
| GMPBackbone=CLIP ViT-B/32, P-Ratio (%)=25, GFLOPs=6.7542026.06 | 6.18 | 1.16 | 4.76 | 0.39 | 0.07 | 100 | 4.75 | 1.06 | |
| OMPBackbone=CLIP ViT-B/32, P-Ratio (%)=30, GFLOPs=5.2582026.06 | 4.14 | 0.91 | 10 | 0.35 | 0.07 | 70 | 2.55 | 0.46 | |
| OMPBackbone=CLIP ViT-B/32, P-Ratio (%)=25, GFLOPs=5.6332026.06 | 3.53 | 0.29 | 0 | 0 | 0 | 100 | 0.56 | 0.13 | |
| OMPBackbone=ActionCLIP ViT-B/16, P-Ratio (%)=20, GFLOPs=19.3262026.06 | 3.36 | 0.52 | 0 | 0 | 0 | 0 | 0 | 0 | |
| OMPBackbone=CLIP ViT-B/32, P-Ratio (%)=20, GFLOPs=6.0092026.06 | 3.07 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | |
| GMPModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=30, GFLOPs=6.6302026.06 | 2.68 | 0.87 | — | — | — | — | — | — | |
| Kwon et al.Model=ActionCLIP ViT-B/32, Pruning Ratio (%)=30, GFLOPs=5.8132026.06 | 2.46 | 0.94 | — | — | — | — | — | — | |
| GMPBackbone=ActionCLIP ViT-B/16, P-Ratio (%)=20, GFLOPs=21.7082026.06 | 2.3 | 1.03 | 44.44 | 0.88 | 0.27 | 55.56 | 2.07 | 0.33 | |
| OMPModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=20, GFLOPs=6.0092026.06 | 2.26 | 1.1 | — | — | — | — | — | — | |
| OMPModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=25, GFLOPs=5.6332026.06 | 2.26 | 1.23 | — | — | — | — | — | — | |
| OMPModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=30, GFLOPs=5.2582026.06 | 2.26 | 1.26 | — | — | — | — | — | — | |
| Kwon et al.Backbone=CLIP ViT-B/16, P-Ratio (%)=25, GFLOPs=18.5822026.06 | 2.2 | 4.33 | 3.45 | 15 | 0.2 | 98.85 | 9.01 | 5.7 | |
| GMPBackbone=ActionCLIP ViT-B/16, P-Ratio (%)=30, GFLOPs=20.7192026.06 | 1.58 | 1.07 | 25 | 0.82 | 0.2 | 0 | 0 | 0 | |
| Kwon et al.Backbone=ActionCLIP ViT-B/16, P-Ratio (%)=30, GFLOPs=17.4632026.06 | 0.49 | 1.33 | 30.77 | 2.16 | 0.53 | 96.15 | 4.73 | 1.66 | |
| GMPModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=25, GFLOPs=6.7552026.06 | 0.42 | 1 | — | — | — | — | — | — | |
| Kwon et al.Backbone=CLIP ViT-B/16, P-Ratio (%)=30, GFLOPs=17.4632026.06 | 0.36 | 0.87 | 0 | 0 | 0 | 100 | 2.77 | 1.13 | |
| OMPBackbone=ActionCLIP ViT-B/16, P-Ratio (%)=25, GFLOPs=18.1172026.06 | 0.23 | 0.52 | 100 | 0.27 | 0.07 | 0 | 0 | 0 | |
| OMPBackbone=ActionCLIP ViT-B/16, P-Ratio (%)=30, GFLOPs=16.9092026.06 | 0.1 | 0.36 | 0 | 0 | 0 | 0 | 0 | 0 | |
| GMPBackbone=ActionCLIP ViT-B/16, P-Ratio (%)=25, GFLOPs=21.2252026.06 | 0.06 | 0.65 | 0 | 0 | 0 | 0 | 0 | 0 | |
| DenseModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=0, GFLOPs=7.5122026.06 | — | — | 27.2 | 36.04 | 9.41 | 96.35 | 41.64 | 33.33 | |
| ECoFLaPModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=20, GFLOPs=5.7602026.06 | — | — | 32.01 | 24.95 | 6.96 | 98.48 | 25.23 | 21.4 | |
| ECoFLaPModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=25, GFLOPs=5.7652026.06 | — | — | 33.74 | 25.51 | 7.49 | 98.51 | 27.8 | 21.87 | |
| ECoFLaPModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=30, GFLOPs=5.4382026.06 | — | — | 32.49 | 15.13 | 4.24 | 98.47 | 15.52 | 12.86 | |
| GMPModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=20, GFLOPs=6.8792026.06 | — | — | 37.74 | 0.54 | 0.2 | 13.21 | 1.07 | 0.07 | |
| GMPModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=25, GFLOPs=6.7552026.06 | — | — | 33.75 | 0.98 | 0.27 | 0 | 0 | 0 | |
| GMPModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=30, GFLOPs=6.6302026.06 | — | — | 20.59 | 0.26 | 0.07 | 60.61 | 2.72 | 0.2 | |
| Kwon et al.Model=ActionCLIP ViT-B/32, Pruning Ratio (%)=20, GFLOPs=6.3802026.06 | — | — | 30.63 | 11.82 | 3.51 | 97.12 | 16.39 | 11.13 | |
| Kwon et al.Model=ActionCLIP ViT-B/32, Pruning Ratio (%)=25, GFLOPs=6.0962026.06 | — | — | 23.02 | 4.06 | 1.19 | 93.62 | 8.3 | 4.84 | |
| Kwon et al.Model=ActionCLIP ViT-B/32, Pruning Ratio (%)=30, GFLOPs=5.8132026.06 | — | — | 9.35 | 0.52 | 0.13 | 85.61 | 3.86 | 1.19 | |
| OMPModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=20, GFLOPs=6.0092026.06 | — | — | 8.05 | 0.39 | 0.07 | 6.06 | 0.99 | 0 | |
| OMPModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=25, GFLOPs=5.6332026.06 | — | — | 31.03 | 1.38 | 0.27 | 0 | 0 | 0 | |
| OMPModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=30, GFLOPs=5.2582026.06 | — | — | 21.51 | 1.19 | 0.2 | 0 | 0 | 0 | |
| Rationale-Informed PruningModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=20, GFLOPs=6.5932026.06 | — | — | 28.64 | 36.42 | 9.68 | 96.86 | 39.74 | 32.74 | |
| Rationale-Informed PruningModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=25, GFLOPs=5.8982026.06 | — | — | 35.97 | 33.34 | 9.68 | 97.77 | 32.09 | 26.31 | |
| Rationale-Informed PruningModel=ActionCLIP ViT-B/32, Pruning Ratio (%)=30, GFLOPs=5.6302026.06 | — | — | 33.25 | 28.75 | 7.89 | 98.31 | 28.05 | 23.33 |