A Benchmark for Inpainting of Clothing Images with Irregular Holes
Accepted to the European Conference on Computer Vision, ECCV 2020, Advances in Image Manipulation workshop and challenges on image and video manipulation AIM2020
Framework: PyTorch
Datasets: FashionGen, FashionAI, DeepFashion, DeepFashion2
Masks: qd-imd
Base code: birdortyedi
Hardware: 2x GTX 1080Ti or 2x Tesla V100
git clone https://github.com/birdortyedi/fashion-image-inpainting.git
- Python >= 3.6
- torch >= 1.2.0
- torchvision >= 0.4.2
- tensorboardX >= 1.9
- numpy >= 1.16
- pillow >= 6.2.0
- h5py >= 2.10.0
- tqdm >= 4.38.0
- colorama >= 0.4.1
python main.py
TODO: use argparser
Hyper-parameter | Value |
---|---|
Optimizer | Adam |
Betas | [0.5, 0.9] |
Generator LR | 0.0002 |
Discriminator LR | 0.0001 |
Scheduler | Exponential |
Schedule Rate | 0.9 |
Batch Size | 32 |
Mask Form | Free-form |
Multi-GPU | True |
Methods | Range |
---|---|
Flipping | Horizontal |
Resizing | 256 |
Please feel free to open an issue or to send an e-mail to [email protected]