Clarification on Camera Annotations, Keypoint Units, and Coordinate System

[safi-harshil]

Dear authors,

Thank you for releasing the InterHand2.6M dataset—it's an incredibly valuable resource for the research community.

While working with the annotations and camera parameters, I encountered a few points that I would greatly appreciate clarification on:

1. Camera Position: In the camera annotation files, what does the "camera position" refer to? Is it a relative position, and if so, relative to what (e.g., world origin, hand root, or another reference)?

2. Units for Intrinsics and Extrinsics: What are the units used in the camera intrinsics and extrinsics files (e.g., pixels, millimeters)?

3. Keypoint Units: What are the units of the annotated keypoint coordinates? Are they directly expressed in those units, or do they require transformation?

4. Coordinate System Convention: Could you please confirm the Cartesian coordinate conventions used? Specifically:

• Is index 0 = X, 1 = Y, 2 = Z?
• Is the origin at the camera center?
• Are the axes defined such that:
  X: left = negative, right = positive
  Y: above = positive, below = negative
  Z: forward = increasing depth?

Clarifying these details would help ensure correct interpretation during downstream processing and model validation.

— Harshil, Research Trainee at IIT Gandhinagar

[mks0601]

1. Camera position in world coordinate system
2. Intrinsic: pixels, extrinsics: millimeter
3. millimeter
4. index: 0=x, 1=y, 2=z. origin: world coordinate system. axis direction: I do not know.