Inquiry Regarding Event Data Representation Parameters #51
Description
Dear @huaaaliu ,
I hope this message finds you well. First of all, I would like to extend my gratitude for the outstanding work and valuable contributions your team has made to the community.
I am reaching out with a query regarding the parameters used in event data representation, specifically concerning the number of time bins utilized.
In the document D. Multi-Modal Data Representations -- (4) RGB-Event, there is a description that contrasts this work with previous studies:
Unlike earlier approaches [26] which convert event data using B = 3 time bins, our method initially embeds events into a voxel grid with an enhanced temporal resolution, setting the upscale factor of the event bin size to 6. Subsequently, every 6 panels are aggregated to generate a detailed event embedding.
However, the ablation study illustrated in Figure 7 indicates optimal performance when B equals 3.
Could you please clarify whether the concepts of these two sets of time bins are directly comparable? For achieving the best results, should one opt for a bin count of 3 or 6 when constructing the voxel grid?
Your guidance on this matter would be greatly appreciated, as it will significantly influence the approach I take in my own research.
Thank you very much for your attention to this query.
Best regards,
pihai