Chainer implementation for Image-to-Image Translation Using Conditional Adversarial Networks
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
list
readme_images
.gitignore
README.md
combine_A_and_B.py
dataset.py
model.py
split_data_sets_cityscapes.py
test.py
train.py
updater.py
utils.py

README.md

Image-to-Image Translation (Chainer)

Chainer implementation for Image-to-Image Translation Using Conditional Adversarial Networks, it's transplanted from pix2pix.

Result

Step by Step

Download Datasets --- Cityscapes

Sign up, Log in & Download gtFine_trainvaltest.zip & leftImg8bit_trainvaltest.zip from cityscapes

Split Dataset into train, val & test set

    python split_data_sets.py --root [Image Folder] --list [List Folder]  

Note: Run python split_data_sets.py -h for more details.

Combine A & B into a single image

    python combine_A_and_B.py --list [List Path] --save_dir [Save Folder]

Generate list file for train, val & test dataset

	ls train > train.txt
	ls val > val.txt
	ls test > test.txt

Note: Run commands above in folder containing train, val & test subfolder

Train

	python train.py

Note: Run python train.py -h for more options.

Test

	python test.py

Note: Run python test.py -h for more options.

Network Architecture

Generator

U-net

Discriminator

PatchGAN, FOV is determined by n_layers.

Acknowledgments

Code borrows heavily from Chainer: DCGAN.