UniCAC

Towards Universal Computational Aberration Correction in Photographic Cameras: A Comprehensive Benchmark Analysis

Xiaolong Qian* · Qi Jiang* · Yao Gao · Lei Sun† · Zhonghua Yi · Kailun Yang · Luc Van Gool · Kaiwei Wang†
* Equal contribution  |  † Corresponding authors
Zhejiang University · INSAIT, Sofia University "St. Kliment Ohridski" · Hunan University

✨ Summary

• Comprehensive CAC benchmark: large-scale dataset and benchmark for photographic cameras, built via automatic optical design to cover diverse optical aberrations.
• Optical Degradation Evaluator (ODE): framework for objectively quantifying optical aberrations and CAC task difficulty.
• Key design insights: systematic comparison of 24 methods reveals three dominant factors for CAC performance — prior utilization, network architecture, and training strategy — and analyzes their respective roles.

📰 News

• 2026-05-22: The CACBench benchmark dataset has been released on Hugging Face.

The released dataset contains four parts:

CACBench_Data_release/
CACBench_Lib_release/
CACBench_Zmx_release/
CACBench_Checker_release/

• CACBench_Data_release/: image data and labels for training, validation, and testing.
• CACBench_Lib_release/: degradation/lens library files used by the dataset.
• CACBench_Zmx_release/: optical design files, including ZMX/ZDA files.
• CACBench_Checker_release/: checker files for evaluation and visualization.

✅ TODO

1. Data & Optical Assets

• Release Train / Val / Test datasets
• Release degradation/lens library files
• Release Zemax files (.zmx, .zda)
• Release simulated PSF assets

2. Evaluation Metrics

• Release checker files for evaluation and visualization
• Release ODE calculation script
• Release Overall Performance (O.P.) evaluation script

3. Unified Framework

• Release unified codebase integrating most of the evaluated baseline models

🎓 Citation

If you find this work useful, please consider citing:

@article{qian2026towards,
  title={Towards Universal Computational Aberration Correction in Photographic Cameras: A Comprehensive Benchmark Analysis},
  author={Qian, Xiaolong and Jiang, Qi and Gao, Yao and Sun, Lei and Yi, Zhonghua and Yang, Kailun and Van Gool, Luc and Wang, Kaiwei},
  journal={arXiv preprint arXiv:2603.12083},
  year={2026}
}