Tensorflow implementation of Automatic Portrait Matting on paper "Automatic Portrait Segmentation for Image Stylization"
- Tensorflow
- scipy
- six
- PIL
- numpy
mkdir Model_zoo and download http://www.vlfeat.org/matconvnet/models/beta16/imagenet-vgg-verydeep-19.mat
download data from http://xiaoyongshen.me/webpage_portrait/index.html
mkdir data and copy the following folders and files into data/:
- images_mask/
- images_data/
- images_data_crop/
- images_tracker/
- portraitFCN_data/
- portraitFCN+_data/
- meanmask.png
- testlist.mat
- trainlist.mat
Some of the folders data above, will be obtained by additional scripting. To obtain additional data, run the data_download.m
and generate_data_file.m
respectively
FCN.py and FCN_puls.py implements the two models in that paper. To train:
if __name__ == "__main__": tf.app.run() #pred()
After train, predict:
if __name__ == "__main__": #tf.app.run() pred()
由于shenxiaoyong网站崩溃,提供数据的百度云下载链接 Link: http://pan.baidu.com/s/1dE14537 Password: ndg8 仅供学术交流,如有引用,请通知原作者。
Thanks for Xiaoyong Shen and his paper
"Deep Automatic Portrait Matting" Xiaoyong Shen, Xin Tao, Hongyun Gao, Chao Zhou, Jiaya Jia. European Conference on Computer Vision (ECCV), 2016
I can't reach the performance in this paper. So please modify it as you want and inform me if any improvement.