ZHENG2049/EMS-YOLO

Offical implementation of "Deep Directly-Trained Spiking Neural Networks for Object Detection" (ICCV2023)

Deep Directly-Trained Spiking Neural Networks for Object Detection (ICCV2023)

Requirements

The code has been tested with pytorch=1.10.1,py=3.8, cuda=11.3, cudnn=8.2.0_0 . The conda environment can be copied directly via environment.yml. Some additional dependencies can be found in the environment.txt.

Install

$ git clone https://github.com/BICLab/EMS-YOLO.git
$ pip install -r requirements.txt

Pretrained Checkpoints

We provide the best and the last trained model based on EMS-Res34 on the COCO dataset.

detect.py runs inference on a variety of sources, downloading models automatically from the COCO_EMS-ResNet34 .

The relevant parameter files are in the runs/train.

Training & Addition

Train

The relevant code for the Gen1 dataset is at /g1-resnet. It needs to be replaced or added to the appropriate root folder.

For gen1 dataset:

python path/to/train_g1.py --weights ***.pt --img 640

For coco dataset:

python train.py

Calculating the spiking rate:

Dependencies can be downloaded from Visualizer.

python calculate_fr.py

Contact Information

@inproceedings{su2023deep,
  title={Deep Directly-Trained Spiking Neural Networks for Object Detection},
  author={Su, Qiaoyi and Chou, Yuhong and Hu, Yifan and Li, Jianing and Mei, Shijie and Zhang, Ziyang and Li, Guoqi},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={6555--6565},
  year={2023}
}

YOLOv3 is a family of object detection architectures and models pretrained on the COCO dataset, and represents Ultralytics open-source research into future vision AI methods, incorporating lessons learned and best practices evolved over thousands of hours of research and development.

Our code is also implemented in this framework, so please remember to cite their work.