A TensorFlow implementation using residual convolution autoencoder with/without discriminator for Face Completion

• Use CELEBA
• Generate noise at center (64x64) in utils.py
• Data path

  -Face-Completion
    -autoencoder
      -data
        -train
        -test
    -autoencoder_gan
      -data
        -train
        -test
  

• Python == 2.7
• Tensorflow == 1.4
• Skimage
• Matplotlib == 2.0.0

• Load pre-trained
  Put checkpoint files under ./autoencoder/model folder and set restore=True

• Train and test

python main.py
--epoch 
--batch_size
--data_path=./data
--model_path=./model
--output_path=./out
--graph_path=./graph
--restore=False
--mode=train/test

• Visualization

tensorboard --logdir=./graph

• Train and test

python main.py
--epoch 
--batch_size
--data_path=./data
--model_path=./model
--output_path=./out
--graph_path=./graph
--restore=False
--mode=train/test

• Visualization

tensorboard --logdir=./graph