ShapeMoire

Pytorch implementation of [ShapeMoire: Channel-Wise Shape-Based Network for Image Demoireing].

Usage

Environment Installation

Requirements
- Linux or macOS (Windows is not currently officially supported)
- Python 3.8
- PyTorch 1.9.0
- CUDA 11.1
- GCC 7.3.0

Install dependencies.

Here is a full script for setting up ShapeMoire with conda.

# build conda environment
conda create -n shapemoire python=3.8 -y
conda activate shapemoire

# install latest PyTorch prebuilt with the default prebuilt CUDA version 
conda install pytorch torchvision -c pytorch

# install other dependencies
conda install lpips==0.1.4 numpy==1.25.2 opencv_python==4.8.0.76 Pillow==10.0.1 PyYAML==6.0.1 skimage==0.0 tensorboardX==2.6.2.2 thop==0.1.1.post2209072238 tqdm==4.66.1

Data Preparation

Download Dataset

You can download four open datasets: FHDMi, TIP2018, UHDM and LCDMoire from the Internet.

Link dataset path under $ShapeMoire/data.

Data Structure

Finally, the total data structure is shown like this:

Shapemoire/
|---configs/
|---data/
|   |---FHDMi/
|   |   |---train/
|   |   |---test/
|   |---TIP2018/
|   |   |---train/
|   |   |---test/
|   |---UHDM/
|   |   |---train/
|   |   |---test/
|   |---LCDMoire/
|   |   |---train/
|   |   |---test/

Modeling

Train

For training process, we use config file in $ShapeMoire/configs to define model, dataset and hyber parameters.

Run the following command to start a training process. You should specify the model and dataset before training.
```
python train_{MODEL_NAME}.py --config {DATASET_NAME}.yaml
```
Note:
- The default config file is defined to train Shapemoire. For training baseline model, modify config with 'TEST_BASELINE: True'.
- For ESDNet, in order to train ESDNet-L, modify config with 'SAM_NUM:2'.
Test

Run the following command to start a testing process.

Except for choosing model and dataset, You need to specify the checkpoint using the parameter 'LOAD_PATH' within config.
```
python test_{MODEL_NAME}.py --config {DATASET_NAME}.yaml
```

Main Results