Skip to content

joanshen0508/Deep-Semantic-Face-Deblurring

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

14 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Deep semantic face deblurring

=======================================================================================

Ziyi Shen, Wei-sheng Lai, Tingfa Xu, Jan Kautz and Ming-Hsuan Yang

Deep Semantic Face Deblurring

IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2018

=======================================================================================

In this paper, we propose to deblur face images using a multiscale network. A face parsing neural network is embedded into the deblurring framework, a multiple loss functions is applied to constraint the model. It is capable of deblurring the face image with more accurate semantic details.

The whole testing datasets have been released on our project website https://sites.google.com/site/ziyishenmi/cvpr18_face_deblur

The blur kernel for training our model is also been released.

If you want test our method on your own face data, please align your face data fisrtly.

Requirements MATLAB (We test with MATLAB R2016a on Windows 10)

Test Pre-trained Models

Compile matconvnet: -> cd matconvnet-1.0-beta22/matlab

-> vl_compilenn('enableGpu', 1, 'enableCudnn', 1)

-> cd ../../

We provide the matconvnet in, ./DL_deblur_net./matconvnet-1.0-beta22./matlab

Run 'main_deblur18.m' to test the example.png

or

You also can run 'main_deblur_all.m' to test the blur images in ./blur_image and ./real_blur_image.

We provide two models here.

  1. net_G_P_S_F_GAN and net_P_P_S_F_GAN Our model with L1 loss + parsing Loss + structure loss +feathure loss

  2. net_G_P_S_F and net_G_P_S_F

Our model with L1 loss + parsing Loss + structure loss + feathure loss + adversarial loss

About

Deep Semantic Face Deblurring

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published