Add DINOv3 OWEED ViTDet training pipeline#5534
Conversation
Add a frozen Hugging Face DINOv3 ViT-B/16 backbone for ViTDet-style Mask R-CNN training while keeping the detector heads and SimpleFeaturePyramid architecture compatible with Detectron2 LazyConfig. Add OWEED-specific training and evaluation configs for the original, tiled v0, and tiled v1 datasets, including bbox-only evaluation for memory-sensitive validation, best-checkpoint tracking by bbox AP, and a mask-eval config for full COCO bbox+segm metrics. Add a COCO tiling utility for offline 1024px tiled datasets with overlap support, plus a full-validation inference script that can dump annotated images, log timing/GPU memory, and optionally export low-memory full-resolution contours scaled from compact ROI masks instead of materializing dense 4K masks on GPU.
|
Hi @NikhilSandy! Thank you for your pull request and welcome to our community. Action RequiredIn order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you. ProcessIn order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA. Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with If you have received this in error or have any questions, please contact us at cla@meta.com. Thanks! |
1 participant
Add a frozen Hugging Face DINOv3 ViT-B/16 backbone for ViTDet-style Mask R-CNN training while keeping the detector heads and SimpleFeaturePyramid architecture compatible with Detectron2 LazyConfig.
Add OWEED-specific training and evaluation configs for the original, tiled v0, and tiled v1 datasets, including bbox-only evaluation for memory-sensitive validation, best-checkpoint tracking by bbox AP, and a mask-eval config for full COCO bbox+segm metrics.
Add a COCO tiling utility for offline 1024px tiled datasets with overlap support, plus a full-validation inference script that can dump annotated images, log timing/GPU memory, and optionally export low-memory full-resolution contours scaled from compact ROI masks instead of materializing dense 4K masks on GPU.
Thanks for your contribution!
If you're sending a large PR (e.g., >100 lines),
please open an issue first about the feature / bug, and indicate how you want to contribute.
We do not always accept features.
See https://detectron2.readthedocs.io/notes/contributing.html#pull-requests about how we handle PRs.
Before submitting a PR, please run
dev/linter.shto lint the code.