TarDAL

Target-aware Dual Adversarial Learning and a Multi-scenario Multi-Modality Benchmark to Fuse Infrared and Visible for Object Detection.

Work have been accepted by CVPR 2022.

The paper and dataset will available soon.

Quick Start Examples

You can try our method online (free) in Colab.

Install

We recommend you to use the conda management environment.

conda create -n tardal python=3.8
conda activate tardal
pip install -r requirements.txt

Fuse or Eval

We offer three pre-trained models.

Name	Description
TarDAL	Optimized for human vision. (Default)
TarDAL+	Optimized for object detection.
TarDAL++	Optimal solution for joint human vision and detection accuracy.

python fuse.py --src data/sample/s1 --dst runs/sample/tardal --weights weights/tardal.pt --color
python fuse.py --src data/sample/s1 --dst runs/sample/tardal+ --weights weights/tardal+.pt --color --eval
python fuse.py --src data/sample/s1 --dst runs/sample/tardal++ --weights weights/tardal++.pt --color --eval

--color will colorize the fused images with corresponding visible color space.