Skip to content

Commit b5c567e

Browse files
Add MoViNet base Kinetics 400 checkpoints.
PiperOrigin-RevId: 422883324
1 parent e8c9e94 commit b5c567e

File tree

1 file changed

+18
-0
lines changed

1 file changed

+18
-0
lines changed

official/projects/movinet/README.md

Lines changed: 18 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -186,6 +186,24 @@ single CPU core. Observed latency may differ depending on hardware
186186
configuration. Measured on a stock Pixel 4 (Android 11) and x86 Intel Xeon
187187
W-2135 CPU.
188188

189+
### Kinetics 400
190+
191+
We also have checkpoints for Kinetics 600 models available. See the Kinetics 600
192+
sections for more details. To load checkpoints, set `num_classes=400`.
193+
194+
#### Base Models
195+
196+
| Model Name | Top-1 Accuracy | Top-5 Accuracy | Input Shape | GFLOPs\* | Checkpoint |
197+
|------------|----------------|----------------|-------------|----------|------------|
198+
| MoViNet-A0-Base | 69.40 | 89.18 | 50 x 172 x 172 | 2.7 | [checkpoint (12 MB)](https://storage.googleapis.com/tf_model_garden/vision/movinet/movinet_a0_base_k400.tar.gz) |
199+
| MoViNet-A1-Base | 74.57 | 92.03 | 50 x 172 x 172 | 6.0 | [checkpoint (18 MB)](https://storage.googleapis.com/tf_model_garden/vision/movinet/movinet_a1_base_k400.tar.gz) |
200+
| MoViNet-A2-Base | 75.91 | 92.63 | 50 x 224 x 224 | 10 | [checkpoint (20 MB)](https://storage.googleapis.com/tf_model_garden/vision/movinet/movinet_a2_base_k400.tar.gz) |
201+
| MoViNet-A3-Base | 79.34 | 94.52 | 120 x 256 x 256 | 57 | [checkpoint (29 MB)](https://storage.googleapis.com/tf_model_garden/vision/movinet/movinet_a3_base_k400.tar.gz) |
202+
| MoViNet-A4-Base | 80.64 | 94.93 | 80 x 290 x 290 | 110 | [checkpoint (44 MB)](https://storage.googleapis.com/tf_model_garden/vision/movinet/movinet_a4_base_k400.tar.gz) |
203+
| MoViNet-A5-Base | 81.39 | 95.06 | 120 x 320 x 320 | 280 | [checkpoint (72 MB)](https://storage.googleapis.com/tf_model_garden/vision/movinet/movinet_a5_base_k400.tar.gz) |
204+
205+
*GFLOPs per video on Kinetics 400.
206+
189207
## Prediction Examples
190208

191209
Please check out our [Colab Notebook](https://colab.research.google.com/github/tensorflow/models/blob/master/official/projects/movinet/movinet_tutorial.ipynb)

0 commit comments

Comments
 (0)