Segmentation Model Support

[pakshi10]

Describe the bug

I was not getting any response on my previous bug related to segmentation model . Is the support for segmentation model is added or not ? If not when it will be added . I am happy to contribute if development is needed for segmentation model .

[juni3227]

I have inspected the code, and it becomes crystal clear that segmentation models have not been implemented

the reason for it is
Segmentation model output and OD model has completely different output list of tensors but they are all handled by class Vec2Box in the yolo/.tils/bounding_box_utils.py

Sementation model's output is ready for bbox generation, where as OD is not, that is because of implementation of MultiHeadSegmentation includes MultiHeadObejectDetection already, but it seems it's output is also processed in that bounding_box_utils.py
That, really should not happen and I could not find any function that calculates seg_loss.

[cmyers009]

Does anyone know about any MIT licensed version of Yolo-v9-S (segmentation)

It would be huge if this could be built.

[Eric-Canas]

Yes,

I was trying to migrate qrdet to this solution, because it is a relieve to have a fully MIT version of YOLO.

But when I tried to train the segmentation model, I got an error when arriving to postprocessing. Because it gets to Vec2Box while the output is completely different :(

It would be amazing if segmentation models could be supported soon.