About testset

[LI-CHIN-CHIA]

Hi, I have a question about the test set.
Why do we still need to put in the original image and the annotated image (mask) during testing?
I tried to put all-black and all-white images into the annotated photos (mask) in the test set. The results would be different from the annotated images (mask), and the boundaries would be somewhat blurry.
I know that adding a mask makes it easier to calculate errors, but if there is a new data set, it still needs to be marked first.

[xianlin7]

Hi, I have a question about the test set. Why do we still need to put in the original image and the annotated image (mask) during testing? I tried to put all-black and all-white images into the annotated photos (mask) in the test set. The results would be different from the annotated images (mask), and the boundaries would be somewhat blurry. I know that adding a mask makes it easier to calculate errors, but if there is a new data set, it still needs to be marked first.

The annotated image is used to generate the prompt. This version of SAMUS is an interactive segmentation model, you must give at least one positive point (in our code, we generate the point prompt from the annotated images) to let the model know what it should segment.

[LI-CHIN-CHIA]

Thanks,another question,if classes are more than two,how to prepare annotated image.