AmineMarnissi / UDAT

The alignment of feature distributions between visible and thermal domains is crucial for achieving effective object detection. Our proposed adaptive detector, integrated into the faster R-CNN architecture, tackles the domain shift problem and enhances overall performance. Experimental evaluations are demonstrated on the KAIST and FLIR ADAS dataset

Feature distribution alignments for object detection in the thermal domain

📖 Feature distribution alignments for object detection in the thermal domain

[Paper]   [Project Page]   [Demo]
Mohamed Amine Marnissi ...

---

🔧 Dependencies and Installation

• Python >= 3.7 (Recommend to use Anaconda or Miniconda)
• PyTorch >= 1.7
• Option: NVIDIA GPU + CUDA
• Option: Linux

Installation

We now provide a clean version of GFPGAN, which does not require customized CUDA extensions.

1. Clone repo

   git clone https://github.com/AmineMarnissi/UDAT.git
   cd UDAT

2. Install dependent packages

   # Create the environment from the environment.yml file:
   conda env create -f environment.yml
   
   # Activate the new environment:
   conda activate UDAT
   
   # Verify that the new environment was installed correctly:
   conda env list

Datasets

• KAIST: Download the Thermal KAIST and Visible KAIST dataset.
• FLIR: Download the Thermal FLIR and Visible FLIR dataset.

Models

Pre-trained Models

In our experiments, we used two pre-trained models on ImageNet, i.e. ResNet50. Please download the model from:

• ResNet50: Dropbox, VT Server

Download and make the model in data/pretrained_model/.

Training

bash kaist_train.sh

bash flir_train.sh

Test

python test_flir.py

Demo

BibTeX

@article{article,
author = {Marnissi, Mohamed and Fradi, Hajer and Sahbani, Anis and ESSOUKRI BEN AMARA, Najoua},
year = {2022},
month = {02},
title = {Feature distribution alignments for object detection in the thermal domain},
journal = {The Visual Computer}
}

📧 Contact

If you have any question, please email [e-mail].