A PyTorch Implmentation of the paper, Image Inpainting for Irregular Holes Using Partial Convolutions. Architecture may not be an exact match of due to the limited description of hyperparameters and architecture details.
See lightning
branch for latest developments in generalising the training and monitoring process.
Trained on Open Images dataset for 5 days. Requires more 4 more days of training and 1 day of fine tuning (around $300 of GCP credit).
Latest model available here.
- Partial Conv: https://github.com/NVIDIA/partialconv
- Keras Implementation: https://github.com/MathiasGruber/PConv-Keras
- Create architecture
- Implement training architecture
- Save the model to an external location
- Document code
- Train for full duration
- Refactor code to use pytorch-lightning
- Create Colab prediction implementation