search

tub-rip / event_based_optical_flow

The official implementation of "Secrets of Event-based Optical Flow" (ECCV2022 Oral and IEEE T-PAMI 2024)
GNU General Public License v3.0
138 stars 13 forks source link

Event mask for evaluation #23

Closed hquan7395 closed 9 months ago

hquan7395 commented 10 months ago

Hi,

Firstly, thank you for your work.

Why is that the evaluation metrics are computed with the mask of the original events instead of the expanded events (due to the 30k requirement)? Do the results in your paper reflect this (both qualitative results and the illustrative figures)?

shiba24 commented 9 months ago

Hi @hquan7395 , the reason of using the original event mask is for a fair comparison with other previous works. In the table we need a comparison that is based on the same calculation step.

However, from my experience, even if I use the expanded events, the metrics results would be similar (haven't tested though).