tum-vision / DEVO

[3DV '24] Deep Event Visual Odometry
MIT License
61 stars 4 forks source link

Question about how event data is transformed into voxels. #3

Closed Windaway closed 4 months ago

Windaway commented 7 months ago

Thank you for your contribution. I have a question about how event data is transformed into voxels. The paper mentions that voxel grids preserve temporal information by bilinear interpolation of event counts in time. It discretizes the time dimension into five bins and normalizes each voxel grid to zero mean and unit variance. However, I thought events themselves have only two states, on or off. How can they be saved as five bins? Or did you accumulate both on and off states, directly summing up +1 and -1? Could you clarify this in advance?

knelk commented 4 months ago

We are using the vid2e code for that, see https://github.com/uzh-rpg/rpg_vid2e

I just added the script for creating the simulated events for tartan air