FEANet-pytorch

This is the official pytorch implementation of FEANet: FEANet: Feature-Enhanced Attention Network for RGB-Thermal Real-time Semantic Segmentation (IEEE IROS). Some of the codes are borrowed from MFNet and RTFNet.

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.

Introduction

Extensive experiments on the urban scene dataset demonstrate that our FEANet outperforms other state-of-the-art (SOTA) RGB-T methods in terms of objective metrics and subjective visual comparison (+2.6% in global mAcc and +0.8% in global mIoU). For the 480 × 640 RGB-T test images, our FEANet can run with a real-time speed on an NVIDIA GeForce RTX 2080 Ti card. Please take a look at thepaper.

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:

FEANet : https://drive.google.com/file/d/1hT4ah8D3wjB1nlUjhSmCEYxFx_vC78ki/view?usp=sharing

python run_own_pth.py -dr [data_dir] -d [test] -f best.pth

Training

python train.py -dr [data_dir] -ls 0.03 -b 5 -em 100

RESULTS

Citation

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

@inproceedings{DBLP:conf/iros/DengFLWYGCHGL21,
  author    = {Fuqin Deng and
               Hua Feng and
               Mingjian Liang and
               Hongmin Wang and
               Yong Yang and
               Yuan Gao and
               Junfeng Chen and
               Junjie Hu and
               Xiyue Guo and
               Tin Lun Lam},
  title     = {FEANet: Feature-Enhanced Attention Network for RGB-Thermal Real-time
               Semantic Segmentation},
  booktitle = {{IEEE/RSJ} International Conference on Intelligent Robots and Systems,
               {IROS} 2021, Prague, Czech Republic, September 27 - Oct. 1, 2021},
  pages     = {4467--4473},
  publisher = {{IEEE}},
  year      = {2021},
  url       = {https://doi.org/10.1109/IROS51168.2021.9636084},
  doi       = {10.1109/IROS51168.2021.9636084},
  timestamp = {Wed, 22 Dec 2021 12:37:50 +0100},
  biburl    = {https://dblp.org/rec/conf/iros/DengFLWYGCHGL21.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}

Demos

Future Work

A High Accuracy benchmark mark will come soon! The mIoU may achieve 60% the first time.

blog

FEANet

Update

Contact

Hua Feng：[email protected]

Mingjian Liang: [email protected]