README.md

Blur Magnitude Estimator(BME)

This task aims to estimate the blur magnitude for each pixel from a blurry image as below example. The details will be in our ECCV 2024 Paper. The downstream task is our DADeblur for video deblurring.

Prepare BME dataset

In this task, we use the RAFT and GoPro to generate training dataset for BME. The details will be in our ECCV 2024 Paper

python generate_dataset/generate_dataset.py

Pretrained Model Weight

You can download the our model weight from this link BME Model Weight

Dataset Structure

dataset/
├── video1/
│ ├── blur_image/
│ └── blur_mag_np/
├── video2/
│ ├── blur_image/
│ └── blur_mag_np/
├── video3/
│ ├── blur_image/
│ └── blur_mag_np/

Train

python main.py --training_dataset_path="your training dataset" --testing_dataset_path="your testing dataset" --weight_path="weight output path"

Inference

python main.py --infer_dataset_path="your inference dataset"  --infer_output_path="your output folder path"  --weight_path="model weight path" --test_only