Action Recognition on THU-EACT-50-CHL
69.85Top-1 AccuracyHoloEv-Net-Base
Evaluation results
| Method | Links | ||
|---|---|---|---|
| HoloEv-Net-BaseInput Type=Holographic, Backbone=ResNet-18, FLOPs=1.9G, Params=12.5M, Latency=9.6ms2026.02 | 69.85 | 79.6 | |
| HoloEv-Net-SmallInput Type=Holographic, Backbone=MobileNetV3-Small, FLOPs=0.1G, Params=4.4M, Latency=7.4ms2026.02 | 67.1 | 78.86 | |
| MVFNetInput Type=Multi-view, Backbone=ResNet50, FLOPs=32.8G, Params=23.6M, Latency=17.5ms2026.02 | 59.56 | 77.49 | |
| MVF-NetInput Type=Multi-view, Backbone=ResNet34&18, FLOPs=5.6G, Params=33.6M, Latency=13.7ms2026.02 | 56.47 | 76.53 | |
| ESTInput Type=Learnable, Backbone=ResNet34, FLOPs=4.2G, Params=21.5M, Latency=11.7ms2026.02 | 55.07 | 76.02 | |
| CoSTInput Type=Multi-view, Backbone=ResNet50, FLOPs=19.2G, Params=25.4M, Latency=21.3ms2026.02 | 51.84 | 75 | |
| R2Plus1DInput Type=Frame, Backbone=ResNet34, FLOPs=81.4G, Params=63.5M, Latency=11.2ms2026.02 | 50.54 | 73.15 | |
| SlowFastInput Type=Frame, Backbone=ResNet50, FLOPs=1.4G, Params=33.6M, Latency=19.6ms2026.02 | 46.14 | 73.16 | |
| TSMInput Type=Frame, Backbone=ResNet50, FLOPs=1.4G, Params=24.3M, Latency=11.8ms2026.02 | 44.3 | 71.31 | |
| I3DInput Type=Frame, Backbone=ResNet50, FLOPs=12.8G, Params=35.3M, Latency=13.1ms2026.02 | 42.41 | 68.49 | |
| C3DInput Type=Frame, Backbone=3D CNN, FLOPs=0.4G, Params=147.2M, Latency=8.4ms2026.02 | 41.54 | 68.57 |