Spectral Transfer Guided Active Domain Adaptation for Thermal Imagery

If you make use of this code, please cite the following paper (and give us a star ✨):

@InProceedings{stgada2023,
    author    = {Ustun, Berkcan and Kaya, Ahmet Kagan and Cakir Ayerden, Ezgi and Altinel, Fazil },
    title     = {Spectral Transfer Guided Active Domain Adaptation for Thermal Imagery},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
    month     = {June},
    year      = {2023},
    pages     = {4322-4331}
}

Overview

This repository contains official implementation of "Spectral Transfer Guided Active Domain Adaptation for Thermal Imagery" paper (accepted to CVPR 2021 Perception Beyond the Visible Spectrum (PBVS) workshop).

Environment

• Python 3.7
• Pytorch 1.8.0
• torchvision 0.9
• Numpy 1.20

Dataset Preparation

• Download FLIR ADAS dataset: Link
• Download MS-COCO dataset:
  • Train images
  • Val images
  • Train/Val annotations
• After you downloaded the datasets, prepare the dataset .txt files similar to the files in 'RGB' folder and put them in that folder.

Running

1. Modify the 'ini.config' file.
2. To train STGADA, run the command below.

(stgada) $ python run.py 

Parameter Name	Type	Definition
[path]	str	Path to the txt files
[source]	str	Name of the Source dataset
[target]	str	Name of the Target dataset
[target_test]	str	Name of the Target Test dataset
[lr]	float	Learning rate
[batch]	int	Batch size
[stgada_lambda]	float	Lambda parameter
[stgada_margin]	float	Margin parameter
[l]	float	l parameter of FDA (the beta parameter)
[device]	int	GPU device ID

• Log

We also provide our experiment logs saved in RGB_result/{dataset_source}_{dataset_target}.log. For example, mscoco_flir.log and the best model along with the models saved after the active epochs.

Acknowledgement

This repo is mostly based on:

• "https://github.com/TencentYoutuResearch/ActiveLearning-SDM"