Optical Flow for Spike Camera with Hierarchical Spatial-Temporal Spike Fusion

Authors: Rui Zhao, Ruiqin Xiong, Jian Zhang, Xinfeng 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 demonstrate that the proposed method achieves state-of-the-art performance compared with the existing methods.
Researcher Affiliation Academia Rui Zhao1,2, Ruiqin Xiong1,2*, Jian Zhang3, Xinfeng Zhang4, Zhaofei Yu1,2,5, Tiejun Huang1,2,5 1National Key Laboratory for Multimedia Information Processing, School of Computer Science, Peking University 2National Engineering Research Center of Visual Technology, School of Computer Science, Peking University 3School of Electronic and Computer Engineering, Peking University 4School of Computer Science and Technology, University of Chinese Academy of Sciences 5Institute for Artificial Intelligence, Peking University ruizhao@stu.pku.edu.cn, {rqxiong, zhangjian.sz, yuzf12, tjhuang}@pku.edu.cn, xfzhang@ucas.ac.cn
Pseudocode No The paper describes the proposed method in text and figures, but does not include a dedicated pseudocode block or algorithm listing.
Open Source Code Yes The source codes are available at https: //github.com/ruizhao26/Hi ST-SFlow.
Open Datasets Yes Datasets. SPIFT (Hu et al. 2022) is a dataset that is designed for the training of spike-based optical flow. ... PHM (Hu et al. 2022) is a dataset that is designed for the evaluation of spike-based optical flow.
Dataset Splits No The paper states, "we use SPIFT as the training set and use PHM as the evaluation set," but does not specify explicit training/validation/test splits (e.g., percentages, sample counts, or cross-validation details) for reproducibility.
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU types, or memory specifications used for running experiments.
Software Dependencies No The paper mentions using an Adam optimizer but does not specify any software dependencies with version numbers (e.g., programming language, libraries, or frameworks with their versions).
Experiment Setup Yes In the experiments, we set the input spike frame number as 25 following the SCFlow (Hu et al. 2022), i.e., T half s = 12. The temporal kernel size and stride of {J1, J2, J3} is {5, 5, 5} and {1, 2, 2}, respectively. ... The number of embed modes as 2. The iteration number of the recurrent optimizer is 12. ... The batch size is set as 6. We use an Adam optimizer (Kingma and Ba 2015) with β1 = 0.9 and β2 = 0.999. The model is trained for 50 epochs. The learning rate is initialized as 1e-4 and scaled by 0.8 every 10 epochs.