Image Segmentation
Transformers
PyTorch
detr
vision

Label explanation

#11
by vittoriacav - opened

What are [LABEL_183, LABEL_184, ..., LABEL_248, LABEL_249]?

I found online that there might be more classes than needed and that those LABEL_* can be ignored. However, when I tested the model on an image, I found that the majority of the segmentation was made up of these LABEL_* classes. For example, LABEL_187 corresponded to the sky, LABEL_184 to some kind of vegetation, LABEL_193 to grass, and LABEL_194 to gravel terrain.

Why is that?

Sign up or log in to comment