search

schuy1er / EWF_official

An official code for "Endpoints Weight Fusion for Class Incremental Semantic Segmentation"
MIT License
25 stars 4 forks source link

Reported results in the paper. #2

Closed adaxidedakaonang closed 1 year ago

adaxidedakaonang commented 1 year ago

Seems your test results are based on the cropped test images. I run the 15-1 setting, and the mIoU should be the 62% instead of 67%. I think by default we should not use the cropped test images for final test.

schuy1er commented 1 year ago

Thanks for the question. It's mainly for the convenience of comparison with the previous work. PLOP, MiB, and RCIL all use cropping of the test image, and we continue to adopt this. It should be noted that the new codebases "ALIFE" and "AWT" currently adopt this as well.