TransferNet: Learning Transferrable Knowledge for Semantic Segmentation with Deep Convolutional Neural Network

Created by Seunghoon Hong, Junhyuk Oh, Honglak Lee and Bohyung Han

Project page: [http://cvlab.postech.ac.kr/research/transfernet/]

Introduction

This repository contains the source code for the semantic segmentation algorithm described in the following paper:

• Seunghoon Hong, Junhyuk Oh, Honglak Lee, Bohyung Han, "Learning Transferrable Knowledge for Semantic Segmentation with Deep Convolutional Neural Network" In IEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016.

@inproceedings{HongOLH2016,
  title={Learning Transferrable Knowledge for Semantic Segmentation with Deep Convolutional Neural Network},
  author={Hong, Seunghoon and Oh, Junhyuk and Lee, Honglak and Han, Bohyung},
  booktitle={Computer Vision and Pattern Recognition (CVPR), 2016 IEEE Conference on},
  year={2016}
}

Pleae refer to our arXiv tech report for details.

Installation

You need to compile the modified Caffe library in this repository. Please consult Caffe installation guide for details. After installing rquired libraries for Caffe, you need to compile both Caffe and its Matlab interface as follows:

cd caffe
make all
make matcaffe

After installing Caffe, you can download datasets, pre-trained models, and other libraries by following script:

setup.sh

Training

Training procedures are composed of two steps, which are implemented in different directories:

• training/1_train_attention : pre-train attention and classification network with image-level class labels.
• training/2_train_segmentation : train entire network including a decoder with pixel-wise class labels.

You can run training with following scripts

cd training
./1_train_attention.sh
./2_train_segmentation.sh

Inference

You can run inference on PASCAL VOC 2012 validatoin images using the trained model as follow:

cd inference
matlab -nodesktop -r run_inference

By default, this script will perform an inference on PASCAL VOC 2012 validation images using the pre-trained model. You may need to modify the code if you want to apply the model to different dataset or use the different models.

Licence

This software is for research purpose only. Check LICENSE file for details.