SAM multi-class semantic segmentation

[vecxoz]

Hello,
Thanks a lot for the great framework!

My task involves multi-class semantic segmentation. There is an excellent tutorial on binary segmentation. What should I change to work with masks containing 10 classes? https://auto.gluon.ai/stable/tutorials/multimodal/image_segmentation/beginner_semantic_seg.html

If I try to set MultiModalPredictor(num_classes=10) I get the following error when calling predictor.fit
RuntimeError: Detected the following values in target: tensor([0.0235, 0.0275, 0.0314, 0.0353, 0.0392]) but expected only the following values [0, 1].

Thanks in advance!

[vecxoz]

Answered here: #3677 (comment)

I cloned the repo and installed from the source. Then trained my model on a V100-16GB. Everything worked smoothly.