Unsupervised Optical Flow Estimation with Dynamic Timing Representation for Spike Camera

Authors: Lujie Xia, Ziluo Ding, Rui Zhao, Jiyuan Zhang, Lei Ma, Zhaofei Yu, Tiejun Huang, Ruiqin Xiong

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments show that our method can predict optical flow from spike streams in different high-speed scenes, including real scenes. For instance, our method achieves 15% and 19% error reduction on PHM dataset compared to the best spike-based work, SCFlow, in t = 10 and t = 20 respectively, using the same settings as in previous works.
Researcher Affiliation Academia Lujie Xia1,2, Ziluo Ding2,3, Rui Zhao1,2, Jiyuan Zhang1,2, Lei Ma5, Zhaofei Yu1,2,4, Tiejun Huang1,2, Ruiqin Xiong1,2 1School of Computer Science, Peking University 2National Engineering Research Center of Visual Technology (NERCVT) 3Beijing Academy of Artificial Intelligence 4Institute for Artificial Intelligence, Peking University 5National Biomedical Imaging Center, College of Future Technology, Peking University {lujie.xia, ziluo, lei.ma, yuzf12, tjhuang, rqxiong}@pku.edu.cn {ruizhao, jyzhang}@stu.pku.edu.cn
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes The source code and dataset are available at https://github.com/Bosserhead/USFlow.
Open Datasets Yes Following the settings in previous works [12], we train our method on SPIFT [12] dataset and evaluate it on PHM [12] and our proposed SSES datasets. [...] The source code and dataset are available at https://github.com/Bosserhead/USFlow.
Dataset Splits No The paper states that training details are in the appendix, but does not provide specific training/validation/test dataset splits (percentages or counts) in the main text. It mentions using PHM and SSES datasets for evaluation, and SSES as a "synthetic validation dataset" but doesn't specify how the main training dataset (SPIFT) was split for training/validation.
Hardware Specification Yes With the lightweight representation module, the computational time of USFlow for inferring optical flow between two timestamps on a 3090 GPU is 90.6ms.
Software Dependencies No The paper does not provide specific software dependencies with version numbers.
Experiment Setup No The paper states "All training details are included in the appendix" and mentions data settings for generating optical flow (t=10, t=20), but it does not provide specific hyperparameter values (e.g., learning rate, batch size, number of epochs, optimizer settings) in the main text.