Skip to content

Latest commit

 

History

History
93 lines (69 loc) · 2.36 KB

README.md

File metadata and controls

93 lines (69 loc) · 2.36 KB
Raw

Cycle GAN

Yet another Cycle GAN implementation in PyTorch.

The purpose of this implementation is Well-structured, reusable and easily understandable.

Prerequisites

  • System

    • Linux or macOS
    • CPU or (NVIDIA GPU + CUDA CuDNN)
      • It can run on Single CPU/GPU or Multi GPUs.
    • Python 3

  • Libraries

    • PyTorch >= 0.3.0
    • Torchvision >= 0.2.0
    • scipy >= 1.0.0
    • Pillow >= 0.2.0

Training

python train.py \
    --data_A_dir=./datasets/apple2orange/trainA \
    --data_B_dir=./datasets/apple2orange/trainB \
    --output_dir=./outputs

If you set test_data_A_dir and test_data_B_dir then generate A->B and B->A when end of every epoch.

python train.py \
    --data_A_dir=./datasets/apple2orange/trainA \
    --data_B_dir=./datasets/apple2orange/trainB \
    --test_data_A_dir=./datasets/apple2orange/testA \
    --test_data_B_dir=./datasets/apple2orange/testB \
    --output_dir=./outputs

Use python train.py --help to see more options.

Transferring

For single file

python transfer.py \
    --model=./outputs/model \
    --src=./datasets/apple2orange/testA/n07740461_41.jpg \
    --out=./outputs/apple2orange.png

For directory

python transfer.py \
    --src_dir=./datasets/apple2orange/testA \
    --out_dir=./outputs/testA

Use python transfer.py --help to see more options.

File structures

network.py and model.py is main implementations.

  • cyclegan
    • config.py : Training options
    • network.py : The neural network architecture of Cycle GAN
    • model.py : Calculate loss and optimizing
    • utils
      • data.py : Utilities for loading data
      • logger.py : Utilities for logging
      • ops.py : Utilities for tensor operations
      • tester.py : Utility functions especially for testing
  • train.py : A script for CycleGAN training
  • transfer.py : A script for transferring with pre-trained model

TODO

  • Visualizing training progress with Visdom
  • Add some nice generated images and videos :-)

References