Skip to content

Latest commit

 

History

History
35 lines (23 loc) · 1.2 KB

README.md

File metadata and controls

35 lines (23 loc) · 1.2 KB

Boundary Equilibrium GAN implementation in PyTorch

Paper is available in arxiv.

Usage

  1. Download Algined and Cropped CelebA dataset.
  2. Use the face_detect script to crop images.
  3. To train the model, run the main script (Check flags for other tunable options):
    python main.py --cuda --base_path . --data_path <data_path> 
  1. To generated interpolated results:
    python main.py --cuda --base_path . --load_step <saved epoch to load> --train 0
  1. To use any pretrained models, just plug in --base_path trained_models --model_name 128 --load step 208000 in the above step for model trained on 128x128 images.

Results

Interpolation on 128x128 images after 206000 epochs:

interpolation128

Interpolation on 64x64 images after 97000 epochs:

interpolation64

Generated 64x64 images after 97k epochs:

128_gens

Generated 128x128 images after 200k epochs:

128_gens