Ahmet Selim Çanakçı

I would expect something that is as easy to work with (kind of similar to `SemanticSegmenterOutput`, maybe like `MaskSegmenterOutput` or `UniversalSegmenterOutput`) as the following script would work easily. ```python from...

@Rocketknight1 , the current implementation only returns `Sam3ImageSegmentationOutput` (you can find the relevant code down below). Forward function definition of `Sam3Model` ```python def forward( self, pixel_values: Optional[torch.FloatTensor] = None, vision_embeds:...

@yonigozlan Thanks for the reply! Unfortunately, I don’t have much spare time to tackle this myself right now as well, but I did a quick skim of the codebase. I...

I agree this is a design decision in which project owner @qubvel should comment on it.

I am not part of the team but I would recommend you to have a look at `dinov3/eval/segmentation/models/__init__.py` there is a method called `build_segmentation_decoder` which builds DinoV3 backbone with Mask2Former...

Did you use Mask2Former head for segmentation? @JohnMBrandt