/Inpainting-Keras

Implement Inpainting techniques in Keras using Tensorflow

Primary LanguagePython

Context Encoder

Based on Context Encoders: Feature Learning by Inpainting [1] and Semi-Supervised Learning with Context-Conditional Generative Adversarial Networks [2]

Requirements

all environment settings are stored in requirements.txt

Summary

This project trains a model to predict a square patch lost of data inside an image. To execute the training the following code should be run.

python context.py

Example

python ccgan.py

Example

NOTE: For the generator a low resolution image is concatenated within the network

This project uses CIFAR10 data with only cats and dogs to reduce number of samples.

References

[1] Context Encoders: Feature Learning by Inpainting

[2] Semi-Supervised Learning with Context-Conditional Generative Adversarial Networks