Awesome

This code repository provides a method for generating pseudo labels for thermal infrared image segmentation using the Segment Anything Model (SAM). The generated dataset can be used to pretrain segmentation models for specific categories with minimal labeled data.

Overview of our proposed method:

<img src="images/index.png" width="400" height="300" alt="overview"/><br/>

Framework of the proposed method:

<img src="images/framework.png" width="1000" height="400" alt="framework"/><br/>

Label generation with masks generated by SAM:

<img src="images/mask2label.png" width="400" height="500" alt="mask2labe;"/><br/>

Some samples of our generated dataset SATIR:

<img src="images/dataset.png" width="1000" height="900" alt="dataset;"/><br/>