/MCHE-CF-for-Multispectral-Pedestrian-Detection

Primary LanguagePythonMIT LicenseMIT

MCHE-CF for Multispectral Pedestrian Detection

Introduction

Multiscale Cross-modal Homogeneity Enhancement and Confidence-aware Fusion for Multispectral Pedestrian Detection

Usage

1. Dependencies

This code is tested on [Ubuntu18.04 LTS，MATLAB R2018b，python 3.7，pytorch 1.5，CUDA 10.1].

make sure the GPU enviroment is the same as above, otherwise you may have to compile the nms and utils according to https://github.com/ruotianluo/pytorch-faster-rcnn.

1. conda activate [your_enviroment]
2. pip install -r requirments.txt

2. Prerequisites

You need to prepare the dataset with the instructions in pytorch-faster-rcnn to prepare KAIST dataset.

3. Pretrained Model

We use VGG16 pretrained models in our experiments. You can download the model from:

Download them and put them into the data/pretrained_model/.

4. Train

Install all the python dependencies using pip:

pip install -r requirements.txt

Compile the cuda dependencies using following simple commands:

cd lib
sh make.sh

Train the dataset using following commands:

python trainval_net.py

You can download the trained model parameters here, (extract code: aaaa) for direct testing on the KAIST dataset.

5. Test

Test the dataset using following commands:

python gen_result.py

The result will generate at './result', then use the matlab code to test the model.

Citation

If you find our work useful in your research, please consider citing:

@inproceedings{MCHE-CF for Multispectral Pedestrian Detection,
    author = {Ruimin Li, Jiajun Xiang, Feixiang Sun, Ye Yuan, Longwu Yuan, Shuiping Gou},
    title = {Multiscale Cross-modal Homogeneity Enhancement and Confidence-aware Fusion for Multispectral Pedestrian Detection},
    booktitle = IEEE Transactions on Multimedia,
    year = {2023}
}