Object Skeleton Detection with FCN

This repo contains codes for our relevant papers on Object Skeleton Detection in Natural Images, they are

  1. Object Skeleton Extraction in Natural Images by Fusing Scale-associated Deep Side Outputs, CVPR2016
  2. DeepSkeleton: Learning Multi-task Scale-associated Deep Side Outputs for Object Skeleton Extraction in Natural Images, IEEE Trans on Image Processing
  3. Hi-Fi: Hierarchical Feature Integration for Skeleton Detection, IJCAI2018

Simple Steps For training:

  1. clone source and build caffe inside;
  2. download SK-LARGE dataset to skeleton/data/ and do data augmentation follow instructions there;
  3. download initial model vgg16convs.caffemodel to skeleton/models/;
  4. python train.py (by default we use the 'FSDS' which is proposed in paper[1]).

Refer to http://kaiz.xyz/deepsk for detailed instructions, frequently asked questions and pretrained models.

FAQ

Frequently asked questions


Copyright KAI ZHAO