Action Recognition (Dense Labeling) on MultiTHUMOS (test)
36.4mAPlatent super-events
Evaluation Results
| Method | Links | |
|---|---|---|
| latent super-eventsbackbone=I3D2017.12 | 36.4 | |
| I3D + Temporal Pyramidpyramid level=32017.12 | 31.2 | |
| I3D + LSTM2017.12 | 29.9 | |
| MultiLSTMFeature=LSTM + i + a + o2015.07 | 29.7 | |
| Predictive-corrective2017.12 | 29.7 | |
| I3D baselinetype=baseline2017.12 | 29.7 | |
| Multi-LSTM2017.12 | 29.6 | |
| LSTM + i + aFeature=input connections + attention2015.07 | 29.1 | |
| LSTM + iFeature=input connections with averaging2015.07 | 28.7 | |
| LSTMImplementation=in the spirit of [4]2015.07 | 28.1 | |
| Two-stream + LSTM2017.12 | 28.1 | |
| Two-stream CNNOptical Flow=single-frame flow2015.07 | 27.6 | |
| Two-stream2017.12 | 27.6 | |
| Single-frame CNNBackbone=VGG-162015.07 | 25.4 | |
| IDT2015.07 | 13.3 |