RTFNet-pytorch

This is the official pytorch implementation of RTFNet: RGB-Thermal Fusion Network for Semantic Segmentation of Urban Scenes (IEEE RA-L). Some of the codes are borrowed from MFNet. Note that our implementations of the evaluation metrics (Acc and IoU) are different from those in MFNet. In addition, we consider the unlabelled class when computing the metrics.

The current version supports Python>=3.8.10, CUDA>=11.3.0 and PyTorch>=1.11.0, but it should work fine with lower versions of CUDA and PyTorch. Please modify the Dockerfile as you want. If you do not use docker, please manually install the dependencies listed in the Dockerfile.

Introduction

RTFNet is a data-fusion network for semantic segmentation using RGB and thermal images. It consists of two encoders and one decoder.

Dataset

The original dataset can be downloaded from the MFNet project page, but you are encouraged to download our preprocessed dataset from here.

Pretrained weights

The weights used in the paper:

RTFNet 50: http://gofile.me/4jm56/9VygmBgPR RTFNet 152: http://gofile.me/4jm56/ODE2fxJKG

Usage

• Assume you have docker and NVIDIA Container Toolkit installed. First, you need to build a docker image. Then, download the dataset:

$ cd ~ 
$ git clone https://github.com/yuxiangsun/RTFNet.git
$ cd ~/RTFNet
$ docker build -t docker_image_rtfnet .
$ mkdir ~/RTFNet/dataset
$ cd ~/RTFNet/dataset
$ (download our preprocessed dataset.zip in this folder)
$ unzip -d .. dataset.zip

• To reproduce our results (for different RTFNet variants, please mannully change num_resnet_layers in RTFNet.py and weight_name in run_demo.py):

$ cd ~/RTFNet
$ mkdir -p ~/RTFNet/weights_backup/RTFNet_50
$ cd ~/RTFNet/weights_backup/RTFNet_50
$ (download the RTFNet_50 weight in this folder)
$ mkdir -p ~/RTFNet/weights_backup/RTFNet_152
$ cd ~/RTFNet/weights_backup/RTFNet_152
$ (download the RTFNet_152 weight in this folder)
$ docker run -it --shm-size 8G -p 1234:6006 --name docker_container_rtfnet --gpus all -v ~/RTFNet:/workspace docker_image_rtfnet
$ (currently, you should be in the docker)
$ cd /workspace
$ python3 run_demo.py

The results will be saved in the ./runs folder.

• To train RTFNet (for different RTFNet variants, please mannully change num_resnet_layers in RTFNet.py):

$ docker run -it --shm-size 8G -p 1234:6006 --name docker_container_rtfnet --gpus all -v ~/RTFNet:/workspace docker_image_rtfnet
$ (currently, you should be in the docker)
$ cd /workspace
$ python3 train.py
$ (fire up another terminal)
$ docker exec -it docker_container_rtfnet bash
$ cd /workspace
$ tensorboard --bind_all --logdir=./runs/tensorboard_log/
$ (fire up your favorite browser with http://localhost:1234, you will see the tensorboard)

The results will be saved in the ./runs folder.

Note: Please change the smoothing factor in the Tensorboard webpage to 0.999, otherwise, you may not find the patterns from the noisy plots. If you have the error docker: Error response from daemon: could not select device driver, please first install NVIDIA Container Toolkit on your computer!

Citation

If you use RTFNet in an academic work, please cite:

@ARTICLE{sun2019rtfnet,
author={Yuxiang Sun and Weixun Zuo and Ming Liu}, 
journal={{IEEE Robotics and Automation Letters}}, 
title={{RTFNet: RGB-Thermal Fusion Network for Semantic Segmentation of Urban Scenes}}, 
year={2019}, 
volume={4}, 
number={3}, 
pages={2576-2583}, 
doi={10.1109/LRA.2019.2904733}, 
ISSN={2377-3766}, 
month={July},}

Demos

About VSCode and Docker

We suggest use VSCode and Docker for deep learning research. Note that this repo already contains the .devcontainer folder, which is needed by VSCode. For more details, please refer to this tutorial.

Contact

[email protected]