mt-cly/fpr

FPR: False Positive Rectification for Weakly Supervised Semantic Segmentation (ICCV 2023)

fpr's Issues

MCTformer settings

Thanks for the awesome work! I have a question about experiment.
Could you provide the parameter settings for the MCTformer model and the choice of feature maps and CAMs ?

Confusion on matching fpr.log with paper result

Hi,

Thank you for providing the code and logs related to the FPR paper. I have been going through the logs, but I'm having trouble matching the precision, recall, and mIoU values reported in the paper with the outputs from the eval_cam function.

It seems like eval_cam was run with multiple thresholds, and each threshold outputs different precision, recall, and mIoU values. Could you please clarify which threshold corresponds to the results reported in the paper?

Thank you for your help!

Request on sharing dataset

Hi,

Thank you for sharing your work. I have a question regarding the dataset used in the paper. The provided link leads to the original dataset with only 2,000+ segmentation annotations. However, the paper mentions an augmented dataset with 10,582 images.

Could you please provide a link to the augmented dataset or instructions on how to obtain it?

Thank you in advance!