Recognizing Ultra-High-Speed Moving Objects with Bio-Inspired Spike Camera

Authors: Junwei Zhao, Shiliang Zhang, Zhaofei Yu, Tiejun Huang

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results show that the proposed method achieves 73.2% accuracy in recognizing 10 classes of ultra-high-speed moving objects, outperforming existing spike-based recognition methods.
Researcher Affiliation Academia Junwei Zhao1,2, Shiliang Zhang1 , Zhaofei Yu1,2 , Tiejun Huang1,2 1National Key Laboratory for Multimedia Information Processing, School of Computer Science, Peking University 2Institute for Artificial Intelligence, Peking University {jwz, slzhang.jdl, yuzf12, tjhuang}@pku.edu.cn
Pseudocode No The paper includes figures illustrating the proposed architecture and modules, but it does not contain any formal pseudocode or algorithm blocks.
Open Source Code Yes Resources will be available at https://github.com/Evin-X/UHSR.
Open Datasets Yes Additionally, this paper contributes an original real-captured spiking recognition dataset consisting of 12,000 ultra-high-speed (equivalent speed > 500 km/h) moving objects. ... Resources will be available at https://github.com/Evin-X/UHSR. ... Besides above methods, this paper contributes a spiking dataset for Ultra-High-Speed object Recognition, named UHSR dataset.
Dataset Splits No Training and testing sets are spilt as a ratio of 5:1. The paper mentions training and testing splits but does not explicitly mention a separate validation set split.
Hardware Specification Yes The framework is implemented in Py Torch and trained on NVIDIA RTX 4090 GPUs.
Software Dependencies No The paper mentions "Py Torch" as the implementation framework but does not specify a version number for PyTorch or any other software dependencies with their versions.
Experiment Setup Yes We adopt the Image Net pre-trained Res Net-18 as the backbone. Model parameters are trained using the SGD optimizer without data augmentation. The initial learning rate is set to 2e-4 with the Lambda LR scheduler. The training process runs for 30 epochs for each dataset with a batch size of 16. Input samples are downsampled to 124 124 using average pooling. θ/λ is set to 1, constraining the value of brightness intensity maps within [0, 1].