search

by-liu / CALS

Code for our method CALS (Class Adaptive Label Smoothing) for network calibration. To Appear at CVPR 2023. Paper: https://arxiv.org/abs/2211.15088
MIT License
21 stars 1 forks source link

Tiny-imagenet ECE of cross entropy #4

Open SoongE opened 2 months ago

SoongE commented 2 months ago

Thanks for your great paper and code.

I have some questions about reproducing paper results.

I ran this code without any changes except the dataset path on tiny-imagenet. However, the ResNet50 cross entropy (baseline) ECE is too high than paper. In my experiment, the ECE of cross entropy is 18.387 without temperature scaling, while the ECE of the paper is 3.73.

I also used your split and the script in README.

I appreciate your attention to this issue. Thank you for your time and support.

SoongE commented 2 months ago

I have attached the config and log. dist_train_rank0.log config.txt