[suggestion] ultralytics yolov11

[monkeycc]

https://github.com/ultralytics/ultralytics

Detect
Segment
Classify
Pose
OBB

[laggui]

Just a note on proper licensing / attribution since ultralytics has an AGPL-3.0 license. I'm not an expert on these things, so correct me if I am wrong:

• I think a re-implementation of the architecture in Burn can be under MIT/Apache
• The pre-trained weights definitely need to be under their AGPL-3.0 license

Or to get a clear response on both, someone could ask the authors.

[monkeycc]

ultralytics/ultralytics#21696 (comment)

Yes—Burn can stay MIT/Apache as long as the core repo does not include or redistribute Ultralytics code or weights; keep any Ultralytics integration as a separate, optional AGPL-3.0 plugin/repo and don’t bundle our pretrained weights. Users who opt in must comply with AGPL-3.0 or hold an Enterprise License; dynamic loading doesn’t change that. For authoritative terms, see the AGPL 3.0 Software License.

[quietlychris]

I think this would tag into the issue opened in #63 as well, with the out-of-the-box option to train a computer vision model that is capable of performing object detection and/or semantic segmentation with Burn. Especially since using the pre-trained weights isn't possible without a license change, it would be nice to be able to support this via training on a COCO-encoded dataset (including something like the competition dataset at https://cocodataset.org/#download), then use a tool like CVAT to produce COCO-encoded archives on datasets of interest for fine-tuning the more generally-trained network.