[CVPR2025] A Physics-Informed Blur Learning Framework for Imaging Systems

📜 Paper   💻 Code   🌐 Project Page

Environment requirements

The codes was tested on Windows 10, with Python and PyTorch. Required packages:

• numpy
• tqdm
• python
• matplotlib
• torch
• torchvision
• pandas
• opencv-python
• pyyaml

File structure

This repository contains codes for OAE(optical aberration estimation).

OAE
|   README.md
|   main_two_step.py
|
|---configs
|   |   lensname.yaml
|
|---sfrmat5_dist
|
|---dataset 
|   |---lensname
|       |   npy
| 
|---input 
|   |   lensname.xlsx
| 
|---model 
|   |   optics_rgb.py
|   |   PSF_mlp.py
| 
|---results 
|
|---utils 
|   |   tools.py
|   |   train.py

/model contains the optical aberration model.

/dataset includes datasets used for training the optical aberration model.

/sfrmat5_dist contains the SFR calculation algorithm, which was downloaded from ISO 12233.

/results stores the results, including the PSF map and PSF comparisons.

Training

To train an aberration learning model from scratch, run main.py. The results will be saved in /results/lensname.