SMPD-mindspore

This is an implementation of "Stabilizing Multispectral Pedestrian Detection with Evidential Hybrid Fusion" based on MindSpore.

This repository aims to provide a comprehensive detection process based on MindSpore for multispectral pedestrian detection, making it the first of its kind. We encourage everyone to collaborate and contribute towards the continuous improvement and maintenance of this code.

Demo

Prerequisites

You can execute this command to create a conda environment for this repository.

conda env create -f smpd_mindspore.yaml

Note: Please go to the official website [link] to install MindSpore for use on Ascend or GPU.

Run

1. You can directly download the model parameters at smpd_mindspore.ckpt.

2. Add the path of model parameters to the variable load_name in the file detector.py.

3. Download data Kaist, and add the path of data to the variable self._pic_path in the file kaist.py.

4. To test on kaist, simply run:

   python detector.py --dataset kaist --net vgg16 --reasonable
   

Note: Due to different implementations of RoIPooling operations, the results may differ slightly from those based on other frameworks. We have tested and found that the results of other operations are strictly consistent.