zwyang6/seco

[CVPR2024] Separate and Conquer: Decoupling Co-occurrence via Decomposition and Representation for Weakly Supervised Semantic Segmentation

seco's Issues

inconsistent results result with the paper

Thanks for sharing the code.
I managed to run your code but the result I get is quite different from the ones in the paper.

-	CAM	val
training result	73.6	71.8
paper result	74.8	74.0

Here is my training log:
train.log

COCO training file

The train_coco.py file only contains the infer code. It seems like the wrong file was uploaded.

fully-supervised counterparts

Hello! Thank you for your interest in our work on WSSS and for your kind words.

When I read your paper, I became curious about Table 5, which compares the fully-supervised counterparts in the paper.
How did you train the fully-supervised counterparts?
Did you experiment by changing pseudo-labels into ground truth masks?

Thanks!