Confidence-aware Fusion using Dempster-Shafer Theory for Multispectral Pedestrian Detection

You can directly click on the link to download our results for drawing the FPPI-MR curve.

You can refer to the environment of MBNet. link

or

You can follow the steps below to configure the environment.

make sure the GPU enviroment is the same as above (cuda10.0,cudnn7.6), otherwise you may have to compile the nms and utils according to https://github.com/endernewton/tf-faster-rcnn. Besides, check the keras version is keras2.1, i find there may be some mistakes if the keras version is higher. To be as simple as possible, I recommend installing the dependencies with Anaconda as follows:

1. conda create -n python36 python=3.6
2. conda activate python36
3. conda install cudatoolkit=10.0
4. conda install cudnn=7.6
5. conda install tensorflow-gpu=1.14
6. conda install keras=2.1
7. conda install opencv
8. python demo.py

Ensure the composition of the folder is as shown in the figure.

+--data
|   +-- kaist_test
|   |   +-- kaist_test_visible
|   |   +-- kaist_test_lwir
+--output
|   +-- resnet_e7_l280.hdf5
+--framework
+--README.md

The specific operation is as follows.

1. Check the CMPD model is available at ./output/resnet_e7_l280.hdf5
2. Enter folder 'framework': cd framework
3. Run the script: python demo.py
4. The detection result is saved at ./result/. (This folder will be created automatically. )

This pipeline is largely built on MBNet. Thank for this great work. Meanwhile, if you encounter any problems during the configuration process, you can check issue of MBNet to see if you can find answers there.