De2Net

Overview of our proposed De2Net . Two main components are embedded into the backbone architecture, i.e., the feature-level Wiener deconvolution layer and the decomposed kernel prediction network.

Dependencies and Installation

• Python >= 3.7 (Recommend to use Anaconda or Miniconda)
• PyTorch >= 1.7
• NVIDIA GPU + CUDA

1. Clone repo

   git clone https://github.com/HyZhu39/De2Net.git

2. Install Dependencies

   cd De2Net
   pip install -r requirements.txt

3. Install BasicSR

   Compile BasicSR without cuda extensions for DCN (Remember to modify the CUDA paths in make.sh and make sure that your GCC version: gcc >= 5)
   

   sh ./make.sh

Dataset Preparation

To grab datasets we used, Please see Dataset.md for details.

Our pretrained models are released on GoogleDrive or BaiduDrive.

Get Started

Test

We provide quick test code with the pretrained model.

1. Download this repo, as well as the datasets and pretrained models from Google Drive or Baidu Drive, and unzip.

2. Modify the paths to dataset and pretrained model in the following yaml files for configuration.

   ./options/test/ZTE_test.yml
   ./options/test/TOLED_test.yml
   ./options/test/POLED_test.yml
   ./options/test/ZTE_test_real_data.yml

3. Run test code for synthetic data of ZTE dataset.

   python -u basicsr/test.py -opt "options/test/ZTE_test.yml" --launcher="none"

4. Run test code for real data of ZTE dataset.

   python -u basicsr/test.py -opt "options/test/ZTE_test_real_data.yml" --launcher="none"

5. Run test code for T-OLED dataset.

   python -u basicsr/test.py -opt "options/test/TOLED_test.yml" --launcher="none"

6. Run test code for P-OLED dataset.

   python -u basicsr/test.py -opt "options/test/POLED_test.yml" --launcher="none"

7. Check out the results in ./results.

Train

All logging files in the training process, e.g., log message, checkpoints, and snapshots, will be saved to ./experiments and ./tb_logger directory.

1. Prepare datasets. Please refer to Dataset Preparation.

2. Modify config files.

   ./options/train/ZTE_train.yml
   ./options/train/TOLED_train.yml
   ./options/train/POLED_train.yml

3. Run training code for three different datasets.

   python -u basicsr/train.py -opt "options/train/ZTE_train.yml" --launcher="none"
   python -u basicsr/train.py -opt "options/train/TOLED_train.yml" --launcher="none"
   python -u basicsr/train.py -opt "options/train/POLED_train.yml" --launcher="none"

Results

Result on *synthetic* data of **ZTE** dataset.

Result on real data of ZTE dataset.

Result on of T-OLED dataset.

Result on of P-OLED dataset.