Xuban's picture
Initial upload: LibreDOMEDETRs-visdrone (Dome-DETR, academic-research-only)
89b8dff verified
|
Raw
History Blame Contribute Delete
3.02 kB
---
license: other
license_name: dome-detr-academic-research-only
license_link: https://huggingface.co/RicePasteM/Dome-DETR/blob/530230620d1f3261a267d462989cddf204cc6e10/README.md
library_name: libreyolo
pipeline_tag: object-detection
tags:
- object-detection
- dome-detr
- tiny-object-detection
- visdrone
- academic-research-only
- non-commercial
---
# LibreDOMEDETRs-visdrone
Dome-DETR S trained on VisDrone (12 output
classes), converted for LibreYOLO at 800x800.
> ## ACADEMIC-RESEARCH-ONLY WEIGHTS
>
> The upstream model card says these weight files are **for academic research
> purposes only**. They are **not covered by LibreYOLO's MIT license** and must
> not be treated as commercially cleared.
>
> The same upstream card also says the project is Apache-2.0, but it has no
> license metadata and links to a LICENSE file that does not exist in the
> weight repository. LibreYOLO's maintainer approved this mirror by treating
> the Apache statement as the redistribution basis and preserving the stricter
> academic-only sentence as the use restriction. This is LibreYOLO's disclosed
> interpretation, not clarification from the Dome-DETR authors. Review the
> pinned upstream card and the [`LICENSE`](./LICENSE) and [`NOTICE`](./NOTICE)
> in this repository before use.
```python
from libreyolo import LibreYOLO
model = LibreYOLO("LibreDOMEDETRs-visdrone.pt")
results = model.predict("image.jpg")
```
## Source
Official upstream weight file:
[`best_ckpts_dome_2026/dome-s-visdrone_converted.pth`](https://huggingface.co/RicePasteM/Dome-DETR/blob/530230620d1f3261a267d462989cddf204cc6e10/best_ckpts_dome_2026/dome-s-visdrone_converted.pth)
- Upstream model revision: `530230620d1f3261a267d462989cddf204cc6e10`
- Source SHA-256: `42874302fd97e17c2a23a46a8ec35d2ea4d2ffe479f0d632c1327fafbf910eb7`
- Converted SHA-256: `01c60b62f042be2893c3a3d32add8b25055aebcdf4bb3098434c1d7a2ba00e18`
The Dome-DETR architecture source is Apache-2.0 at commit
[`2dde3bc1946a3e9fad9abd0612b59fc39bd6b861`](https://github.com/RicePasteM/Dome-DETR/commit/2dde3bc1946a3e9fad9abd0612b59fc39bd6b861).
Copyright (c) 2025 The Dome-DETR Authors, as stated in its source headers.
That source-code license and LibreYOLO's MIT code license do not remove the
academic-only restriction stated for these pretrained weights.
## Modifications
LibreYOLO adds checkpoint-schema metadata, including the dataset variant,
class names, pinned source revision, and source SHA-256. State-dict keys and
learned tensors are unchanged. The converted checkpoint loads strictly into
LibreYOLO's native implementation. See
[`weights/convert_domedetr_weights.py`](https://github.com/LibreYOLO/libreyolo/blob/dev/weights/convert_domedetr_weights.py).
## License and terms
**Academic research purposes only**, following the stricter statement on the
upstream weight card. The upstream Apache-2.0 statement and its missing weight
repository LICENSE file are documented without omission. See
[`LICENSE`](./LICENSE) and [`NOTICE`](./NOTICE).