search

zhaozhengChen / ReCAM

The official code of CVPR 2022 paper (Class Re-Activation Maps for Weakly-Supervised Semantic Segmentation).
141 stars 18 forks source link

mIoU after irnet #17

Closed YangjiaqiDig closed 2 years ago

YangjiaqiDig commented 2 years ago

is that possible the mIoU after irnet train and random walk would be lower than mIoU in first ReCAM? I worked on my own dataset, but get lower ACC after the irn step (make_sem_seg_labels)

zhaozhengChen commented 2 years ago

I think it shouldn't be if you set reasonable thresholds to train irnet and inference.

YangjiaqiDig commented 2 years ago

I think it shouldn't be if you set reasonable thresholds to train irnet and inference.

okay, thanks. Just too slow to tune the params for irn step