Request for camera distortion parameters / calibration details for ActionNet RGB-D videos
Hi Fourier ActionNet team,
We are using the ActionNet RGB-D videos for geometric projection and camera-frame annotation.
From the released dataset and viewer code, we found a dataset-level pinhole camera calibration, including intrinsics and extrinsics. This calibration appears visually reasonable near the image center. However, projections become noticeably less accurate near the image borders, and straight structures near the border appear curved in the RGB videos. This suggests that lens distortion is significant but not currently available in the released calibration metadata.
Could you clarify or provide:
- The distortion model and coefficients used for the released RGB videos, if available.
- Whether the released RGB videos are raw distorted camera frames, undistorted frames, rectified frames, or cropped/resized outputs.
- Whether the depth videos are aligned to the RGB camera frame, and whether they use the same intrinsics/distortion assumptions.
- Whether camera calibration is shared globally across the dataset or differs by robot / camera / episode.
These details would be very helpful for accurate 3D-2D projection, point-cloud reconstruction, and camera-frame label generation.
Thanks!