Why using cls labels to generate CAMs at inference time? Is it valid?

[Siyuan-Zhou]

At val / test time, in infer_SEAM.py (line 79 to line 82), you use GT cls labels to choose CAMs of these categories and save these specified CAMs as .npy files. I am wondering whether using GT cls labels at inference time is valid in weakly-supervised semantic segmentation. Could you provide me with some hints? Much thanks!

[YudeWang]

Hi @Siyuan-Zhou , Because the generated is pseudo labels on train set. There is another retrain step to train segmentation model on these pseudo labels in fully supervised manner.

[Siyuan-Zhou]

@YudeWang Thanks for your quick reply! I exactly know that you use your generated pseudo segmentation masks to train a segmentation model separately. However, what I am talking about is that in the inference stage you use ground truth category-level labels (i.e. cls labels) to select CAMs, see infer_SEAM.py (line 79 to line 82). I am wondering whether ground truth category-level labels (i.e. cls labels) can be used during inference.

[bityangke]

@Siyuan-Zhou it does inference on the train set not the test set

[Siyuan-Zhou]

@bityangke Thanks.