Action Recognition on Something-Something (val)
51.6Top-1 AccuracyR(2+1)D-152
Evaluation results
| Method | Links | ||
|---|---|---|---|
| R(2+1)D-152Pre-training=IG-Kinetics, Short edge scaling=true, Input type=RGB2019.05 | 51.6 | 78.8 | |
| R(2+1)D-152Pre-training=IG-Kinetics, Input type=RGB2019.05 | 51 | 79 | |
| R(2+1)D-34Pre-training=IG-Kinetics, Short edge scaling=true, Input type=RGB2019.05 | 49.9 | 77.5 | |
| R(2+1)D-34Pre-training=IG-Kinetics, Input type=RGB2019.05 | 49.7 | 77.5 | |
| ECOEnLiteInput type=RGB + flow2019.05 | 49.5 | — | |
| S3D-GInput type=RGB2019.05 | 48.2 | 78.7 | |
| ECOEnLiteInput type=RGB2019.05 | 46.4 | — | |
| NL I3D + Joint GCNInput type=RGB2019.05 | 46.1 | 76.8 | |
| R(2+1)D-34Pre-training=Sports-1M, Input type=RGB2019.05 | 45.7 | 74.5 | |
| R(2+1)D-34Pre-training=Kinetics, Input type=RGB2019.05 | 45.2 | 74.1 | |
| MFNet-C101K (number of training segments)=102018.07 | 43.92 | 73.12 | |
| MFNet-C50K (number of training segments)=102018.07 | 40.3 | 70.93 | |
| MFNet-S50K (number of training segments)=102018.07 | 39.83 | 70.19 | |
| MFNet-C50K (number of training segments)=72018.07 | 37.31 | 67.23 | |
| MFNet-S50K (number of training segments)=72018.07 | 37.09 | 67.78 | |
| MultiScale TRNevaluation_strategy=10-crop2018.07 | 34.44 | 63.2 | |
| MultiScale TRN2018.07 | 33.01 | 61.27 | |
| Pre-3D CNN + Avg2018.07 | 11.5 | 30 |