Domain Adaptive Thermal Object Detection with Unbiased Granularity Alignment

---

RGB-to-RGB Benchmark：

Cityscape-to-Foggycityscape branch

Requirements

• Ubuntu 18.04.5 LTS
• Python 3.6
• CUDA 10.0
• PyTorch 1.0.0
• Faster R-CNN

dataset download

We also provide the download URL of the dataset in the future.

Dataste Download

Compile the code

Compile the cuda dependencies using following simple commands following Faster R-CNN:

cd lib
python setup.py build develop

Pre-trained Models

• ResNet101: Dropbox VT Server

Training and Test

Train the model

sh ./train_scripts/train_flir.sh

Test the well-trained model:

python test_scripts/test_flir.py

📝Related repos

Our project references the codes in the following repos:

• Megvii-Nanjing, CR-DA-DET

other code :

• https://github.com/tiancity-NJU/da-faster-rcnn-PyTorch
• https://github.com/MCG-NJU/TIA/
• https://github.com/harsh-99/SCL
• https://github.com/AmineMarnissi/UDAT

if you have any questions , please contact me at [email protected]

@article{10.1145/3665892,
author = {Shi, Caijuan and Zheng, Yuanfan and Chen, Zhen},
title = {Domain Adaptive Thermal Object Detection with Unbiased Granularity Alignment},
year = {2024},
doi = {10.1145/3665892},
journal = {ACM Trans. Multimedia Comput. Commun. Appl.},
}