Awesome
MambaST
MambaST: A Plug-and-Play Cross-Spectral Spatial-Temporal Fuser for Efficient Pedestrian Detection
Usage
-
Build Docker Image
cd docker bash build.sh
Note: Make sure to change
USERNAME
andUSER_UID
in the script. -
Run Docker Container
cd .. bash docker/run.sh
Note: Update
your_username
andpath_to_project_directory
accordingly. -
Install Third Party Libraries
cd thirdparty/Vim pip install -e mamba-1p1p1 cd ../.. pip install causal-conv1d==1.1.0
-
Download and Prepare Dataset
- Download the KAIST-CVPR15 dataset from here and place it in your dataset directory.
- Copy the dataset:
cp -r ~/dataset/* path_to_your_dataset_directory
- Copy the sanitized annotations format:
cp sanitized_annotations_format_all path_to_your_dataset_directory
-
Update Annotation Path
- Modify
KAIST_ANNOTATION_PATH
inutils/datasets_vid.py
to the absolute path ofsanitized_annotations_format_all
.
- Modify
-
Training
bash train.sh
Sample training code.
-
Testing
bash test.sh
Citation
If you found repo useful, feel free to cite.
@article{gao2024mambast,
title={MambaST: A Plug-and-Play Cross-Spectral Spatial-Temporal Fuser for Efficient Pedestrian Detection},
author={Gao, Xiangbo and Kanu-Asiegbu, Asiegbu Miracle and Du, Xiaoxiao},
journal={arXiv preprint arXiv:2408.01037},
year={2024}
}