Object Skeleton Detection with FCN
This repo contains codes for our relevant papers on Object Skeleton Detection in Natural Images, they are
- Object Skeleton Extraction in Natural Images by Fusing Scale-associated Deep Side Outputs, CVPR2016
- DeepSkeleton: Learning Multi-task Scale-associated Deep Side Outputs for Object Skeleton Extraction in Natural Images, IEEE Trans on Image Processing
- Hi-Fi: Hierarchical Feature Integration for Skeleton Detection, IJCAI2018
Simple Steps For training:
- clone source and build caffe inside;
- download SK-LARGE dataset to
skeleton/data/and do data augmentation follow instructions there;
- download initial model vgg16convs.caffemodel to
python train.py(by default we use the 'FSDS' which is proposed in paper).
Refer to http://kaiz.xyz/deepsk for detailed instructions, frequently asked questions and pretrained models.
Copyright KAI ZHAO