LASNet

This project provides the code and results for 'RGB-T Semantic Segmentation with Location, Activation, and Sharpening', IEEE TCSVT, 2023. IEEE link and arxiv link Homepage

Requirements

python 3.7/3.8 + pytorch 1.9.0 (biult on EGFNet)

Segmentation maps and performance

We provide segmentation maps on MFNet dataset and PST900 dataset under './model/'.

Performace on MFNet dataset

Performace on PST900 dataset

Training

Install 'apex'.
Download MFNet dataset (code: 3b9o) or PST900 dataset (code: mp2h).
Use 'generate_binary_labels.m' to get binary labels, and use 'generate_bound_or_edge.m' to get edge labels.
Run train_LASNet.py (default to MFNet Dataset).

Note: our main model is under './toolbox/models/LASNet.py'

Pre-trained model and testing

Download the following pre-trained model and put it under './model/'. model_MFNet.pth (code: 5th1) model_PST900.pth (code: okdq)
Rename the name of the pre-trained model to 'model.pth', and then run test_LASNet.py (default to MFNet Dataset).

Citation

    @ARTICLE{Li_2023_LASNet,
            author = {Gongyang Li and Yike Wang and Zhi Liu and Xinpeng Zhang and Dan Zeng},
            title = {RGB-T Semantic Segmentation with Location, Activation, and Sharpening},
            journal = {IEEE Transactions on Circuits and Systems for Video Technology},
            year = {2023},
            volume = {33},
            number = {3},
            pages = {1223-1235},
            month = {Mar.},
            }

If you encounter any problems with the code, want to report bugs, etc.