Video-level classification on Kinetics 400 (test)
91.1AccuracyMTV-H
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MTV-HEvaluation protocol=Fully finetuned2023.02 | 91.1 | — | |
| ir-CSN-152* + OracleClassifier=ir-CSN-152, Pre-training=Sports1M, Clip Selector=Oracle2019.04 | 89 | — | |
| CoCa#param=2.1B, #data=4.8B, tuning=FT, resolution=5762022.06 | 88.9 | — | |
| CoCaEvaluation protocol=Frozen backbone, Spatial features=pre-pool, Resolution=Higher2023.02 | 88 | — | |
| ViT-22BEvaluation protocol=Frozen backbone2023.02 | 88 | — | |
| Florence#param=647M, #data=900M, tuning=FT, resolution=3842022.06 | 86.5 | — | |
| ViT-eEvaluation protocol=Frozen backbone2023.02 | 86.5 | — | |
| R(2+1)D-18 + OracleClassifier=R(2+1)D-18, Clip Selector=Oracle2019.04 | 85.4 | — | |
| ViViT-L#param=>307M, #data=300M, tuning=FT2022.06 | 84.9 | — | |
| Uni-Perceiver-L + Conditional MoEs#param=303M, #data=44.1M, tuning=FT 100%2022.06 | 84.2 | — | |
| R3D-18 + OracleClassifier=R3D-18, Clip Selector=Oracle2019.04 | 83 | — | |
| Uni-Perceiver-L + Conditional MoEs#param=303M, #data=44.1M, tuning=PT 1%2022.06 | 83 | — | |
| R(2+1)D-34* + OracleClassifier=R(2+1)D-34, Pre-training=Sports1M, Clip Selector=Oracle2019.04 | 82.9 | — | |
| VATT-L#param=306.1M, #data=238M, tuning=FT, resolution=3202022.06 | 82.1 | — | |
| Uni-Perceiver-L + Conditional MoEs#param=303M, #data=44.1M, tuning=WT2022.06 | 82.1 | — | |
| MC3-18 + OracleClassifier=MC3-18, Clip Selector=Oracle2019.04 | 82 | — | |
| Uni-Perceiver-L#param=303M, #data=44.1M, tuning=FT 100%2022.06 | 81.9 | — | |
| ViViT-L#param=>307M, #data=14.2M, tuning=FT2022.06 | 81.7 | — | |
| I3D-RGB** + OracleClassifier=I3D-RGB, Pre-training=ImageNet, 3D-inflated, Clip Selector=Oracle2019.04 | 81.2 | — | |
| TimeSformer-B#param=121.4M, #data=14.2M, tuning=FT2022.06 | 80.7 | — | |
| ir-CSN-152* + SCSamplerClassifier=ir-CSN-152, Pre-training=Sports1M, Clip Selector=SCSampler2019.04 | 80.2 | 1.6 | |
| Uni-Perceiver-L#param=303M, #data=44.1M, tuning=PT 1%2022.06 | 80 | — | |
| VATT-B#param=87.9M, #data=238M, tuning=FT, resolution=3202022.06 | 79.6 | — | |
| Uni-Perceiver-L#param=303M, #data=44.1M, tuning=WT2022.06 | 79.5 | — | |
| Uni-Perceiver-B + Conditional MoEs#param=86M, #data=44.1M, tuning=FT 100%2022.06 | 79.3 | — | |
| ir-CSN-152* + EmpiricalClassifier=ir-CSN-152, Pre-training=Sports1M, Clip Selector=Empirical2019.04 | 79.2 | 1.5 | |
| ir-CSN-152* + DenseClassifier=ir-CSN-152, Pre-training=Sports1M, Clip Selector=Dense2019.04 | 78.8 | 3 | |
| ir-CSN-152* + UniformClassifier=ir-CSN-152, Pre-training=Sports1M, Clip Selector=Uniform2019.04 | 78.5 | 1.5 | |
| ir-CSN-152* + RandomClassifier=ir-CSN-152, Pre-training=Sports1M, Clip Selector=Random2019.04 | 77.8 | 1.5 | |
| Uni-Perceiver-B#param=86M, #data=44.1M, tuning=FT 100%2022.06 | 77.7 | — | |
| Uni-Perceiver-B + Conditional MoEs#param=86M, #data=44.1M, tuning=PT 1%2022.06 | 77.2 | — | |
| Uni-Perceiver-B + Conditional MoEs#param=86M, #data=44.1M, tuning=WT2022.06 | 76.8 | — | |
| R(2+1)D-34* + SCSamplerClassifier=R(2+1)D-34, Pre-training=Sports1M, Clip Selector=SCSampler2019.04 | 76.7 | 1.6 | |
| I3D-RGB** + SCSamplerClassifier=I3D-RGB, Pre-training=ImageNet, 3D-inflated, Clip Selector=SCSampler2019.04 | 75.1 | 1.5 | |
| Uni-Perceiver-B#param=86M, #data=44.1M, tuning=PT 1%2022.06 | 74.8 | — | |
| Uni-Perceiver-B#param=86M, #data=44.1M, tuning=WT2022.06 | 74.5 | — | |
| R(2+1)D-34* + EmpiricalClassifier=R(2+1)D-34, Pre-training=Sports1M, Clip Selector=Empirical2019.04 | 74.1 | 1.5 | |
| R(2+1)D-34* + DenseClassifier=R(2+1)D-34, Pre-training=Sports1M, Clip Selector=Dense2019.04 | 74.1 | 3.1 | |
| R(2+1)D-34* + UniformClassifier=R(2+1)D-34, Pre-training=Sports1M, Clip Selector=Uniform2019.04 | 74 | 1.5 | |
| R(2+1)D-34* + RandomClassifier=R(2+1)D-34, Pre-training=Sports1M, Clip Selector=Random2019.04 | 73.8 | 1.5 | |
| I3D-RGB** + DenseClassifier=I3D-RGB, Pre-training=ImageNet, 3D-inflated, Clip Selector=Dense2019.04 | 72.8 | 2.9 | |
| I3D-RGB** + RandomClassifier=I3D-RGB, Pre-training=ImageNet, 3D-inflated, Clip Selector=Random2019.04 | 71.9 | 1.3 | |
| I3D-RGB** + EmpiricalClassifier=I3D-RGB, Pre-training=ImageNet, 3D-inflated, Clip Selector=Empirical2019.04 | 71.9 | 1.3 | |
| I3D-RGB** + UniformClassifier=I3D-RGB, Pre-training=ImageNet, 3D-inflated, Clip Selector=Uniform2019.04 | 71.8 | 1.3 | |
| R(2+1)D-18 + SCSamplerClassifier=R(2+1)D-18, Clip Selector=SCSampler2019.04 | 70.9 | 1.6 | |
| R(2+1)D-18 + DenseClassifier=R(2+1)D-18, Clip Selector=Dense2019.04 | 68 | 2.4 | |
| R3D-18 + SCSamplerClassifier=R3D-18, Clip Selector=SCSampler2019.04 | 67.3 | 1.6 | |
| MC3-18 + SCSamplerClassifier=MC3-18, Clip Selector=SCSampler2019.04 | 67 | 1.5 | |
| R(2+1)D-18 + EmpiricalClassifier=R(2+1)D-18, Clip Selector=Empirical2019.04 | 66.3 | 1.4 | |
| R(2+1)D-18 + UniformClassifier=R(2+1)D-18, Clip Selector=Uniform2019.04 | 66.2 | 1.4 | |
| R(2+1)D-18 + RandomClassifier=R(2+1)D-18, Clip Selector=Random2019.04 | 65.9 | 1.4 | |
| R3D-18 + DenseClassifier=R3D-18, Clip Selector=Dense2019.04 | 65.2 | 2.4 | |
| MC3-18 + DenseClassifier=MC3-18, Clip Selector=Dense2019.04 | 65.1 | 2.3 | |
| R3D-18 + EmpiricalClassifier=R3D-18, Clip Selector=Empirical2019.04 | 64 | 1.3 | |
| R3D-18 + UniformClassifier=R3D-18, Clip Selector=Uniform2019.04 | 63.8 | 1.3 | |
| MC3-18 + EmpiricalClassifier=MC3-18, Clip Selector=Empirical2019.04 | 63.6 | 1.3 | |
| R3D-18 + RandomClassifier=R3D-18, Clip Selector=Random2019.04 | 63.6 | 1.3 | |
| MC3-18 + UniformClassifier=MC3-18, Clip Selector=Uniform2019.04 | 63.4 | 1.3 | |
| MC3-18 + RandomClassifier=MC3-18, Clip Selector=Random2019.04 | 63 | 1.3 |