ASNet

This project provides the code and results for 'ASNet: An adaptive scene-aware network for RGB-thermal urban scene semantic segmentation'.

Thank you for your interest. The complete training scripts and details will be made publicly available upon publication.

Usage

Requirements

1. Python 3.7.15
2. PyTorch 1.11.0
3. CUDA 11.3+

and pip install -r requirements.txt

Note: For a detailed environment configuration tutorial you can refer to our CSDN blog.

Datasets

dataset

• mfnet/
  • RGB/
    • name1.png
    • name2.png
  • TH/
    • ……
  • Labels/
    • ……
• PST900/
  • ……

If you would like to download the dataset, please contact me.

Pretrain weights:

Download the pretrained Resnet-152 here.

Config

The parameters of the dataset and the network can be modified through this config file.

Results

Results on MFNet (9-class):

Backbone	Modal	mAcc	mIoU
Resnet-152	RGB-T	77.8	55.3

Results on PST900 (5-class):

Backbone	Modal	mAcc	mIoU
Resnet-152	RGB-T	82.85	77.08

All code will be made public after the paper is published.