# HiFill-Pytorch-256

**Repository Path**: ljd132/HiFill-Pytorch-256

## Basic Information

- **Project Name**: HiFill-Pytorch-256
- **Description**: A Unofficial implement of the paper "HiFill" CVPR2021 HuaWei
- **Primary Language**: Python
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 1
- **Forks**: 0
- **Created**: 2023-07-14
- **Last Updated**: 2023-07-14

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# High Resolution Image Inpainting Based on GAN
Unofficial Pytorch Re-implementation of "<a href="https://arxiv.org/abs/2005.09704">Contextual Residual Aggregation for Ultra High-Resolution Image Inpainting</a>"(CVPR 2020 Oral).

The code is based on implementation of <a href="https://github.com/zhaoyuzhi/deepfillv2">deepfillv2</a>. Thanks for the great job.

The project is still in progress, please feel free to contact me if there is any problem.

## Implementation
Besides Contextual Residual Aggregation(CRA) and Light-Weight GatedConvolution in the paper, also add Residual network structure, SN-PatchGAN in this project.
### Preparing
Before running, please ensure the environment is `Python 3.6` and `PyTorch 1.2.0`.

Dataset: Download <a href="http://places2.csail.mit.edu/download.html">Places365-Standard</a> for Training and Testing.

### Training
```bash
python train.py     --epochs 40
                    --lr_g 0.0001
                    --batch_size 4
                    --lambda_perceptual 100
                    --lambda_l1 300 [feel free to change during training]
                    --baseroot [the path of training set]
                    --mask_type 'free_form' [or 'single_bbox' or 'bbox']
                    --imgsize 512
```
```bash
if you have more than one GPU, please change following codes:
python train.py     --multi_gpu True
                    --gpu_ids [the ids of your multi-GPUs]
```
Default training process uses hinge loss as the D_loss, also provide Wgan-GP in the code.

For input size of 512x512 and GPU with memory of 11GB, recommended batchsize is 4.

To change other training parameters, please modify train.py. You can also design your own mask through dataset.py

### Testing
In progress......
### Pre-trained model
Still in training......
### Sample images
To do
### Acknowledgement & Reference

[1] https://github.com/zhaoyuzhi/deepfillv2

```bash
@misc{yi2020contextual,
    title={Contextual Residual Aggregation for Ultra High-Resolution Image Inpainting},
    author={Zili Yi and Qiang Tang and Shekoofeh Azizi and Daesik Jang and Zhan Xu},
    year={2020},
    eprint={2005.09704},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}
```
```bash
@inproceedings{yu2019free,
  title={Free-form image inpainting with gated convolution},
  author={Yu, Jiahui and Lin, Zhe and Yang, Jimei and Shen, Xiaohui and Lu, Xin and Huang, Thomas S},
  booktitle={Proceedings of the IEEE International Conference on Computer Vision},
  pages={4471--4480},
  year={2019}
}
```