SpikSSD: Better Extraction and Fusion for Object Detection with Spiking Neuron Networks

This is the official implementation of the 'SpikSSD: Better Extraction and Fusion for Object Detection with Spiking Neuron Networks' .

Required Data

To evaluate or train SpikSSD you will need to download the datasets:

Dataset Name	Link
PASCAL VOC Dataset	Download PASCAL VOC Detection Dataset
COCO Dataset	Download COCO Detection Dataset
GEN1 Dataset	Download Gen1 Automotive Detection Dataset

Training

Training for SpikSSD

SpikSSD-S

python object_detection.py -num_workers 4 -test -save_ckpt -backbone mdsresnet-18 -b 32 -augmentation -fusion -decode spiking

SpikSSD-L

python object_detection.py -num_workers 4 -test -save_ckpt -backbone mdsresnet-34 -b 32 -augmentation -fusion -decode spiking

Evaluation

During evaluation, it is necessary to substitute the relevant pretrained model data into the appropriate root directory.

Evaluation for SpikSSD

SpikSSD-S

python object_detection.py -num_workers 4 -test -save_ckpt -backbone mdsresnet-18 -b 32 -augmentation -fusion -decode spiking -pretrained path_to_model -no_train

SpikSSD-L

python object_detection.py -num_workers 4 -test -save_ckpt -backbone mdsresnet-34 -b 32 -augmentation -fusion -decode spiking -pretrained path_to_model -no_train

Code Acknowledgments

This code is based on SFOD: Spiking Fusion Object Detector . Thanks to the contributors of SFOD: Spiking Fusion Object Detector .

@inproceedings{fan2024sfod,
  title={SFOD: Spiking Fusion Object Detector},
  author={Fan, Yimeng and Zhang, Wei and Liu, Changsong and Li, Mingyang and Lu, Wenrui},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={17191--17200},
  year={2024}
}