| --- | |
| library_name: onnx | |
| tags: | |
| - action-recognition | |
| - mediapipe | |
| - onnx | |
| - sdk-bundle | |
| # VideoInference | |
| ActionLinkPerception SDK model bundle. | |
| ## Version | |
| - Tag: `curated-v1-multihead-w10-cuda-20260610` | |
| - Source run: `runs/20260610_curated_v1_multihead_w10_cuda` | |
| - Window size: 10 frames | |
| - Dataset: `dataset/clip_data_back_v1_curated_20260610` | |
| ## Files | |
| - `manifest.json`: SDK bundle entry point | |
| - `model.onnx`: ONNX graph | |
| - `action_classifier_curated_v1_multihead_w10_cuda.onnx.data`: ONNX external tensor data | |
| - `model.json`: label maps, heads, keypoints and metadata | |
| - `smoothing.json`: SDK label smoothing configuration | |
| - `reports/`: training log and confusion matrices | |
| ## Usage | |
| Download this repo with Hugging Face CLI and point the SDK model bundle directory to the downloaded folder. | |
| ```powershell | |
| hf download Zihanj/VideoInference --local-dir .\models\VideoInference | |
| ``` | |