What intrinsics should I use

#9
by YOLO2431 - opened

Hi GenRobot Team,

Thank you for sharing the GenRobot dataset — it’s been very helpful. We’re currently working with the MCAP files and have a few clarification questions regarding camera intrinsics:

  1. Resolution scaling
    The camera_info lists 640×480 for calib_w and calib_h, while the videos are 1600×1296. Should we scale fx, fy, cx, cy to match the native resolution? Are the distortion coefficients valid at 1600×1296, or do they require adjustment?

  2. Missing camera_info
    In the first 5000 episodes, camera_info is missing in ~64% (robot0) and ~58% (robot1). Is this expected? Were there specific recording conditions where it wasn’t logged?

  3. Intrinsics variability
    We observe notable variation in fx across episodes (≈247–301). Should we use per-episode intrinsics when available, or is there a recommended canonical set? Is this due to hardware differences or recalibration?

We appreciate any guidance you can provide.

Sign up or log in to comment