This is pytorch implementation of paper "Glow: Generative Flow with Invertible 1x1 Convolutions". Most modules are adapted from the offical TensorFlow version openai/glow.

• Glow model. The model is coded as described in original paper, some functions are adapted from offical TF version. Most modules are tested.
• Trainer, builder and hparams loaded from json.
• Infer after training
• Test LU_decomposed 1x1 conv2d

• Train a model with

  train.py <hparams> <dataset> <dataset_root>
  

• Generate z_delta and manipulate attributes with

  infer_celeba.py <hparams> <dataset_root> <z_dir>
  

Currently, I trained model for 45,000 batches with hparams/celeba.json using CelebA dataset. In short, I trained with follwing parameters

• Download pre-trained model from Dropbox

Following are some samples at training phase. Row 1: reconstructed, Row 2: original.

Use the method decribed in paper to calculate z_pos and z_neg for a given attribute. And z_delta = z_pos - z_neg is the direction to manipulate the original image.

• manipulate Smiling (from negative to positive):

  

• manipulate Young (from negative to positive):

  

• manipulate Pale_Skin (from negative to positive):

  

• manipulate Male (from negative to positive):

  

There might be some errors in my codes. Please help me to figure out.