Floor reference and BVH position channels for SAMP retargeting

#2
by yebifeng - opened

Hello HiPHI authors,

We are retargeting released HiPHI motions to a fixed SAMP character for research. We have checked your shared RH/Y-up human/object world, cm-to-m conversion and synchronized object times; no new motion archive is needed.

Could you confirm:

  1. For the five priority motions below, what floor plane (point/normal in meters, or horizontal height g) was used by your quality evaluation? Please distinguish capture-measured calibration, an evaluation convention, and an estimate. A shared session value plus motion-to-session mapping is sufficient. If g=0 is simply the evaluation convention, that clarification is useful too.
  2. The static preview uses each joint's XYZ positions while the viewer keeps nonroot OFFSET and only animates root positions. Which is recommended for training/evaluation, and which release/code version? In our selected 13 motions the lower-body positions agree within 1 micrometer, but upper-body positions differ by centimeters.
  3. Is there a small released floor/evaluation script or contact-landmark definition supporting those fields? We are not requesting private data or additional large motion files.

Priority motions (received BVH SHA-256):

  • Change_direction-bear_0125 / A033 / d77e04a783f194eff4f642acadaff0fe300844de28070db6e8b6fad1e577f02f
  • Posture-shrink_0025 / A025 / 8f490b7e21db0277d52e5587e40efc79dd94c2b62431d14fe35fe6d5e6588e11
  • Change_posture-stand_up_0016 / A008 / 069e7d3d6e42fabef1498cc64f6f28d2c935db2ea527e9e726a9a9682b844bac
  • Change_posture-stand_up_0019 / A008 / a4f5ba05f6bd116fc25f8a0dd922d796732e15a279451b1354cd8495fcea4c4b
  • Change_posture-stand_up_0020 / A008 / cc53a8cff483c5aba46e9ec382551a8679ed9e78f14f63f3fcf568062620edb6

We have not used first-frame viewer grounding as measured floor, and have not trained on these candidates. Your format's common human/object world is already acknowledged; an additional transform is only needed if these released motions are exceptions.

Thank you.

Noitom Robotics org

Hi @yebifeng ,

Thank you for your interest in HiPHI. Our evaluation uses Y = 0 as the ground-plane convention, and we recommend using all per-joint position and rotation channels to reconstruct the released BVHs.

Best regards,
Jiahao

Sign up or log in to comment