| --- |
| library_name: onnx |
| tags: |
| - object-detection |
| - yolo |
| - yolox |
| - yolo9 |
| - rfdetr |
| - onnxruntime-web |
| - webgpu |
| - browser |
| - libreyolo |
| license: mit |
| --- |
| |
| # LibreYOLO Web — Model Zoo |
|
|
| Pre-exported ONNX models for [libreyolo-web](https://github.com/xuban-ceccon/libreyolo-web), the browser-based multi-family YOLO detection library. |
|
|
| ## Usage |
|
|
| ```bash |
| npm install libreyolo-web onnxruntime-web |
| ``` |
|
|
| ```typescript |
| import { loadModel } from 'libreyolo-web'; |
| |
| // Auto-downloads from this repo, zero config |
| const model = await loadModel('yolox-nano'); |
| const result = await model.predict(imageElement); |
| ``` |
|
|
| ## Available Models |
|
|
| ### YOLOX (letterbox + BGR + 0-255) |
| | Name | Input | Size | File | |
| |------|-------|------|------| |
| | `yolox-nano` | 416 | 3.6MB | `yolox_n.onnx` | |
| | `yolox-tiny` | 416 | 19MB | `yolox_t.onnx` | |
| | `yolox-small` | 640 | 34MB | `yolox_s.onnx` | |
| | `yolox-medium` | 640 | 97MB | `yolox_m.onnx` | |
| | `yolox-large` | 640 | 207MB | `yolox_l.onnx` | |
| | `yolox-xlarge` | 640 | 378MB | `yolox_x.onnx` | |
|
|
| ### YOLO9 (resize + RGB + /255) |
| | Name | Input | Size | File | |
| |------|-------|------|------| |
| | `yolo9-tiny` | 640 | 8MB | `yolo9_t.onnx` | |
| | `yolo9-small` | 640 | 28MB | `yolo9_s.onnx` | |
| | `yolo9-medium` | 640 | 77MB | `yolo9_m.onnx` | |
| | `yolo9-compact` | 640 | 97MB | `yolo9_c.onnx` | |
|
|
| ### RF-DETR (resize + ImageNet norm, no NMS) |
| | Name | Input | Size | File | |
| |------|-------|------|------| |
| | `rfdetr-nano` | 384 | 103MB | `rfdetr_n.onnx` | |
| | `rfdetr-small` | 512 | 109MB | `rfdetr_s.onnx` | |
| | `rfdetr-medium` | 576 | 115MB | `rfdetr_m.onnx` | |
| | `rfdetr-large` | 704 | 116MB | `rfdetr_l.onnx` | |
|
|
| ## Training |
|
|
| All models were trained on COCO using [libreyolo](https://github.com/LibreYOLO/libreyolo) (Python) and exported to ONNX with `model.export(format='onnx', simplify=True)`. |
|
|
| ## License |
|
|
| MIT |
|
|
