Skip to content

xiaolonw/ss-gan

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

68 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ss-gan

This code is developed based on eyescream project with Torch: project site.

This code is the implementation of training and testing for S^2-GAN for the following paper:

Xiaolong Wang and Abhinav Gupta. Generative Image Modeling using Style and Structure Adversarial Networks. Proc. of European Conference on Computer Vision (ECCV), 2016. pdf

BibTeX:

@inproceedings{Wang_SSGAN2016,
    Author = {Xiaolong Wang and Abhinav Gupta},
    Title = {Generative Image Modeling using Style and Structure Adversarial Networks},
    Booktitle = {ECCV},
    Year = {2016},
}

Models and Datasets

The trained models can be downloaded from dropbox.

The pre-processed dataset (NYUv2) including rgb images and TV-denoised Surface Normals in jpgs can be downloaded from dropbox.

The list of training files dropbox.

General Instructions for using the code

For training, one need to:

	Update the path_dataset = '/scratch/xiaolonw/render_data/' in dataset.lua 
	Update the opt.save in train.lua for saving models 

For testing, one can download the models into the ssgan_models folder.

Structure-GAN

The code for Stucture-GAN is in structure-gan:

	train.lua: training Stucture-GAN
	test.lua: testing Stucture-GAN
	ssgan_models/Structure_GAN.net is our trained model

Style-GAN

The code for Style-GAN without FCN constraints is in style-gan-nofcn:

	train.lua: training Style-GAN
	test.lua: testing Style-GAN (To run this you need to download the dataset)
	ssgan_models/Style_GAN_nofcn.net is our trained model

The code for Style-GAN with FCN constraints is in style-gan-fcn:

	train_fcn.lua: training FCN for surface normal estimation
	test_fcn.lua: testing FCN for surface normal estimation (To run this you need to download the dataset)
	ssgan_models/FCN.net is our trained model

	train_gan.lua: training Style-GAN
	test_gan.lua: testing Style-GAN (To run this you need to download the dataset)
	ssgan_models/joint_Style_GAN.net is our trained model

Joint Learning for S^2-GAN

The code for joint learning is in joint-ssgan:

	train.lua: joint learning 
	test.lua: testing S^2-GAN
	ssgan_models/joint_SSGAN.net is our trained model

About

Style and Structure GAN (ECCV 2016)

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages