apocaliss92's picture
feat: plateDetection/yolox-tiny-plate (open-lpr YOLOX-tiny, Apache-2.0)
81fc931 verified
|
Raw History Blame Contribute Delete
1.24 kB
# Source of plateDetection/yolox-tiny-plate
| | |
| --- | --- |
| Upstream repository | https://huggingface.co/faisalthaheem/open-lpr-models |
| Upstream tag / commit | `detector-2026.10.1` / `aed0df35ec2891c5a90437041ebeadb592cc516b` |
| Upstream file | `plate_yolox_tiny_640.onnx`, 20183226 bytes, sha256 `e5350bb2152eb44fa129b2438d1e39bbbf3c27accfe51cd893ce6e17d028d0b9` |
| Upstream LICENSE | sha256 `c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4` (copied here verbatim) |
| Upstream NOTICE | sha256 `8fe7b8450e295ba3a1aae6359cee0cbe69142d2a8b8a395b4ddfb28ad8154fd2` (copied here verbatim) |
| Weights licence | Apache-2.0 |
| Training data | not published; terms UNKNOWN (see README.md) |
Rebuild every file in this folder from the pinned upstream file:
```
python3 scripts/build-yolox-plate-model.py <work-dir>
```
in the CamStack repository (https://github.com/camstack/server). The script
downloads the upstream ONNX by commit and refuses it if the sha256 differs,
applies the graph edit described in README.md, converts the edited ONNX to
OpenVINO IR (fp16 weights) and CoreML (fp16) with CamStack's `convert_lib.py`,
and checks that the edit is exact and that the three runtimes return the same
detections.