README.md

July 10, 2025 · View on GitHub

Towards Adversarial Robustness via Debiased High-Confidence Logit Alignment

📌 Accepted at ICCV 2025

This repository contains the official PyTorch implementation of our ICCV 2025 paper: "Towards Adversarial Robustness via Debiased High-Confidence Logit Alignment".

🏋️‍♂️ Training

Modify the training configuration:

configs_train.yml

Start training:

python train.py

🧪 Robustness Evaluation

Edit the testing configuration:

configs_test.yml

Launch evaluation:

python test_robust.py

📄 Citation

@inproceedings{zhang2025dhat,
  title     = {Towards Adversarial Robustness via Debiased High-Confidence Logit Alignment},
  author    = {Kejia Zhang and Juanjuan Weng and Shaozi Li and Zhiming Luo},
  booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
  year      = {2025}
}