Spatio-temporal Action Detection on JHMDB 14 (3 splits)
77.9Frame-mAP@0.5S3D-G
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| S3D-GGFLOPS=>155.5, Backbone=3D, Tubelet length (K)=7, Fusion type=two-stream2020.01 | 77.9 | — | 80.1 | — | — | |
| MOCGFLOPS=29.4, Pre-training=UCF101-24 [29], Backbone=2D, Tubelet length (K)=7, Fusion type=two-stream2020.01 | 74 | 80.7 | 80.5 | 75 | 60.2 | |
| I3DGFLOPS=>91.0, Backbone=3D, Tubelet length (K)=7, Fusion type=two-stream2020.01 | 73.3 | — | 78.6 | — | — | |
| MOCGFLOPS=29.4, Pre-training=COCO [19], Backbone=2D, Tubelet length (K)=7, Fusion type=two-stream2020.01 | 70.8 | 77.3 | 77.2 | 71.7 | 59.1 | |
| MOCGFLOPS=29.4, Pre-training=ImageNet [3], Backbone=2D, Tubelet length (K)=7, Fusion type=two-stream2020.01 | 68 | 76.2 | 75.4 | 68.5 | 54 |