Request for pipeline or guidance to use EvDeblurVSR with DAVIS346 grayscale data (no GT), Custom Data

[kadirkilicnet]

Hi, thank you for releasing EvDeblurVSR — it’s excellent work!

I’d like to run inference using my own DAVIS346 data, but I’m unsure how to convert it into the required .h5 format expected by EvBlurVSRTestDataset.

My data:
Grayscale frames as frame_00000.png, frame_00001.png, ...

frames.csv: frame_id, timestamp (in microseconds)

events.csv: x, y, timestamp, polarity (0/1)

No ground truth (HR) available

Request:
Could you provide a conversion pipeline or recommendation to generate:

lqs from the grayscale frames

vExpo, vFwd, vBwd voxel grids from the events aligned with frame timestamps

A valid .h5 format and meta_info.txt file (in test-only setup)

This would help us evaluate EvDeblurVSR on real DAVIS event data. We’re happy to contribute back a cleaned script if needed.

TL;DR: I have DAVIS346 grayscale images + events (no GT), and want to test EvDeblurVSR. Could you provide a pipeline or guidance to convert my data (frames + events.csv) into the .h5 format with lqs, vFwd, vBwd, vExpo needed by EvBlurVSRTestDataset?

Thanks!