Action Classification on Kinetics Full (val)
77.7Top-1 AccuracyNL I3D
Evaluation results
| Method | Links | ||
|---|---|---|---|
| NL I3DInputs=RGB, Backbone=ResNet-101, Pre-train=ImNet2017.12 | 77.7 | 93.3 | |
| SANInputs=RGB+Flow+Audio, Backbone=Inception-ResNet-v2, Pre-train=ImNet2017.12 | 77.7 | 93.2 | |
| S3D-GInputs=RGB+Flow, Backbone=Inception, Pre-train=ImNet2017.12 | 77.2 | 93 | |
| R(2+1)DInputs=RGB+Flow, Backbone=ResNet-34, Pre-train=Sports-1M2017.12 | 75.4 | 91.9 | |
| S3D-GInputs=RGB, Backbone=Inception, Pre-train=ImNet2017.12 | 74.7 | 93.4 | |
| I3DInputs=RGB+Flow, Backbone=Inception, Pre-train=ImNet2017.12 | 74.1 | 91.6 | |
| TSNInputs=RGB+Flow, Backbone=Inception, Pre-train=ImNet2017.12 | 73.9 | 91.1 | |
| ARTNetInputs=RGB+Flow, Backbone=ResNet-18, Pre-train=ImNet2017.12 | 72.4 | 90.4 | |
| I3DInputs=RGB, Backbone=Inception, Pre-train=ImNet2017.12 | 71.1 | 89.3 | |
| S3D-GInputs=Flow, Backbone=Inception, Pre-train=ImNet2017.12 | 68 | 87.6 | |
| I3DInputs=Flow, Backbone=Inception, Pre-train=ImNet2017.12 | 63.9 | 85 |