PyTorch reimplementation of Deep Skeleton: Learning Multi-task Scale-associated Deep Side Outputs (LMSDS) for object skeleton extraction in natural images using Python 3.6 and Pytorch 1.0.1
The main objective is to reproduce the algorithm as it is done in the official implementation using Pytorch.
This repository contains two executable files:
- main.py: Training scheme.
- test_results.py: Code to generate test images.
Download SK-LARGE from the author's page and use their code to perform data augmentation.
Once you have your augmented data, you can try training the algorithm.
First you need to download the VGG16 pretrained model.
mkdir model
wget https://download.pytorch.org/models/vgg16-397923af.pth
mv vgg16-397923af.pth model/vgg16.pth
python main.py
Should you wish to test the algorithm:
python test_results.py
Use the skeval code to preform test.
- DeepSkeleton: Learning Multi-task Scale-associated Deep Side Outputs for Object Skeleton Extraction in Natural Images Shen, Wei and Zhao, Kai and Jiang, Yuan and Wang, Yan and Bai, Xiang and Yuille, Alan, IEEE Transactions on Image Processing, 2017, pp. 5298-5311.