Encoders

Convolutional Encoder

Tensorflow implementation of DeepConv Encoder as presented in the paper - Photoshop 2.0: Generative Adversarial Networks for Photo Editing.

Prerequisites

• Python 3.3+
• Tensorflow 1.6
• pillow (PIL)
• matplotlib
• (Optional) CelebA Face Database: Large Scale Face database used for training the model

Usage

To train the Encoder model:

> python train_encoder.py --data_path celeba --input_fname_ppatern .jpg --dcgan_model_dir dcgan_model --encoder_dir encoder_model

• data_path: Directory of the database folder having training images
• input_fname_pattern: Glob pattern of training images
• dcgan_model_dir: DCGAN model directory having model weights
• encoder_dir: Encoder directory to save checkpoints

To compute attribute encodings:

> python average.py --data_path celeba --input_fname_ppatern .jpg --encoder_dir encoder_model --attribute_file attr_labels.txt --attr_encoding_file attr_embed.txt

• data_path: Directory of the database folder having training images
• input_fname_pattern: Glob pattern of training images
• encoder_dir: Encoder directory to save checkpoints
• attribute_file: Label file of the database
• attr_encoding_file: File to save attribute encodings

To manipulate images:

> python manipulate.py --data_path celeba --input_fname_ppatern .jpg --encoder_dir encoder_model --attribute_file attr_labels.txt --attr_encoding_file attr_embed.txt

Results

Results of encoder model after 6th epoch.

Orignal Image - Generated Image (after encoding the image)

License

This project is licensed under the MIT License - see the LICENSE file for details

Author

Abhishek Tandon/ @Tandon-A