# OSNet Weights for Trackers

Pre-trained OSNet x1.0 weights for use with the [trackers](https://github.com/roboflow/trackers) library.

## Files

| File | Description | Source |
|---|---|---|
| `osnet_x1_0_imagenet.pth` | OSNet x1.0 pretrained on ImageNet (general features) | [torchreid](https://github.com/KaiyangZhou/deep-person-reid) |
| `sports_model.pth.tar-60` | OSNet x1.0 fine-tuned on SportsMOT crops for person ReID | [Deep-EIoU](https://github.com/hsiangwei0903/Deep-EIoU) |

## Usage

```python
from trackers.appearance.osnet import OSNetExtractor

# ImageNet pretrained (general)
extractor = OSNetExtractor(weights="imagenet")

# SportsMOT fine-tuned (sports ReID — used by Deep HM-SORT paper)
extractor = OSNetExtractor(weights="sportsmot")
```

## References

- Zhou et al., *Omni-Scale Feature Learning for Person Re-Identification*, ICCV 2019
- Huang et al., *Deep-EIoU*, arXiv:2306.13074
- Gran-Henriksen et al., *Deep HM-SORT*, arXiv:2406.12081