FlareReal600, the first real paired nighttime flare removal dataset, consists of 600 real-paired flare images with 2k/4k resolution.
- 2024.05.20: We released our FlareReal600 dataset (4k resolution) and 500 real-captured flare patterns.
- 2024.01.19: We are holding the MIPI 2024 flare removal challenge (CVPR2024).
- 2024.01.15: We released the training codes and our FlareReal600 dataset (2k resolution).
-
Clone the repo
git clone https://github.com/Zdafeng/FlareReal600.git
-
Install dependent packages
cd FlareReal600 pip install -r requirements.txt -
Install FlareReal600
Please run the following commands in the FlareReal600 root path to install FlareReal600:python setup.py develop
| Baidu Netdisk | Google Drive | Number | Description | |
|---|---|---|---|---|
| FlareReal600(train) | link | link | 600 | FlareReal600 training dataset contains 600 real aligned training images in 2k resolution. |
| FlareReal600(val) | link | link | 50 | FlareReal600 validation dataset contains 50 real aligned training images in 2k resolution. |
| FlareReal600(train) | link | link | 600 | FlareReal600 training dataset contains 600 real aligned training images in 4k resolution. |
| FlareReal600(val) | link | link | 50 | FlareReal600 validation dataset contains 50 real aligned training images in 4k resolution. |
| Flare Only | link | link | 500 | We offers 500 real-captured flare patterns. |
To estimate the flare-free images with trained model, you can run the test.py by using:
python test.py --input dataset/input --output result/ --model_path experiments/net_g_last.pth
To calculate different metrics with trained model, you can run the evaluate.py by using:
python evaluate.py --input result/ --gt dataset/gt/
Training with single GPU
To train a model with FlareReal600 dataset, you can edit the options/uformer_paired_baseline_option.yml and run the following codes.
python basicsr/train.py -opt options/uformer_paired_baseline_option.yml
If you want to use Flare7K++ as additional dataset for training, please use:
python basicsr/train.py -opt options/uformer_paired_flare7kpp_baseline_option.yml
Training with multiple GPU
You can run the following command for the multiple GPU tranining:
CUDA_VISIBLE_DEVICES=0,1 bash scripts/dist_train.sh 2 options/uformer_paired_baseline_option.yml
If you want to use Flare7K++ as additional dataset for training, please use:
CUDA_VISIBLE_DEVICES=0,1 bash scripts/dist_train.sh 2 options/uformer_paired_flare7kpp_baseline_option.yml
If you have any question, please feel free to reach me out at dfeng.zhang@samsung.com.
This project is mainly based on Flare7K.