https://github.com/facebookresearch/dinov3
DINOv3 came out a few weeks ago. Utilizing it as a backbone during training instead of DINOv2 + additional exploration should yield even better performance on synthetic datasets like C3VD and clinical datasets like Colon10K, EndoMaper, etc.
I unfortunately do not have consistent access to spare GPUs I can use toward this effort, but have left this open as a reminder to myself (or in case someone else wishes to try this out and report back).
https://github.com/facebookresearch/dinov3
DINOv3 came out a few weeks ago. Utilizing it as a backbone during training instead of DINOv2 + additional exploration should yield even better performance on synthetic datasets like C3VD and clinical datasets like Colon10K, EndoMaper, etc.
I unfortunately do not have consistent access to spare GPUs I can use toward this effort, but have left this open as a reminder to myself (or in case someone else wishes to try this out and report back).