Image Inpainting using Generative Adversarial Network

---

About

A pytorch implementation of the paper Free Form Image Inpainting with Gated Convolution. The code can be install as a python package called inpaint. The inpaint package provides APIs to :

• Create a Pytorch Dataset for Places365 dataset

  from inpaint.data import PlacesDataset 
  

• Configure and setup Generator and Discriminator

  from inpaint.core.discriminator import PatchDiscriminator
  from inpaint.core.generator import GatedGenerator'
  

• Configure and setup training, evaluation and prediction

  from inpaint.tools import Trainer, Evaluate, predict
  

Tutorial and documentation for APIs are provided in the examples directory.

Results

---

Setup

Here's how to set up inpaint for local development and testing.

1. Install Miniconda

2. Clone the repo locally::

   $ git clone https://github.com/prajnan93/image-inpainting

3. Create a Conda virtual environment using the env.yml file. Install your local copy of the package into the environment::

   • $ conda env create -f env.yml
   • $ conda activate inpaint
   • $ python setup.py develop

4. Please note this repo is not accepting any contributions.

---

Tutorials and Documentation

• Setup the inpaint conda environment as mentioned above.
• Follow the instructions provided in the jupyter notebooks in the directory examples.
• Each notebook in the examples directory provides an example of mask visualization, training, evaluation and prediction.
• Make sure to have at least 16Gb of CUDA GPU memory for training the model. Few example training scripts with differnt model configurations are provided in the scripts directory.

---

Checkpoints of Pretrained models and Predictions in Google Colab

Download pretrained checkpoint .

google colab demo: coming soon

---

Acknowledgements

This repository acknowleges the official implementation of DeepFillV2 Free Form Image Inpainting and the Places365 Dataset datasets.

And thanks to Northeastern University Discovery HPC for providing the compute support.

---

Citations

@article{yu2018generative,
  title={Generative Image Inpainting with Contextual Attention},
  author={Yu, Jiahui and Lin, Zhe and Yang, Jimei and Shen, Xiaohui and Lu, Xin and Huang, Thomas S},
  journal={arXiv preprint arXiv:1801.07892},
  year={2018}
}

@article{yu2018free,
  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},
  journal={arXiv preprint arXiv:1806.03589},
  year={2018}
}

@article{zhou2017places,
  title={Places: A 10 million Image Database for Scene Recognition},
  author={Zhou, Bolei and Lapedriza, Agata and Khosla, Aditya and Oliva, Aude and Torralba, Antonio},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year={2017},
  publisher={IEEE}
}

---

References

• Free-Form Image Inpainting with Gated Convolution

• GatedConvolution_pytorch

• DeepFillv2_Pytorch

• Places365 Dataset

---

License

This python package is for education and research purposes only.

Shield:

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.