Questions about dataset acquisition details

[ryhara]

@kelvin34501 @lixiny @oakink

Hello,

Thank you for providing great dataset.

I have a few questions regarding the data acquisition and processing:

1. Did you use "Motive" for motion capture? If so, could you specify the version used?

2. For the hand marker layout, did you refer to this configuration?
   https://docs.optitrack.com/markersets/hand-and-fingers/active-finger-markerset

3. This marker layout seems to result in relatively low tracking accuracy. (in my experience)
   Did you implement any specific calibration methods, camera arrangements, or recording settings to improve accuracy?

4. Did "data cleaning" contribute significantly to improving the data quality? How long did the data cleaning process take?

5. When you mention "MoCap post-processing software" in Supp, could you clarify which specific software you used?

This information would be very helpful for reproducing the setup or understanding the data more thoroughly.
Looking forward to your response.

[kelvin34501]

Thank you for your interest in OakInk2 dataset!

Answers to your questions:

1. Yes, Motive 3.0 Body.
2. Yes. There are small adaptations to move some markers to more visible positions for better tracking quality.
3. No. MoCap-related settings are all configured in the Motive software. We use this makerset out of a balance in marker numbers and tracking precision and accuracy, as in complex tasks the reflective markers in the capture volume are growing rapidly and too many markers lead to degraded performance. You can use more markers (maybe ~14 - ~19, determined by the number of markers on each finger).
4. Yes. The time of data cleaning is not measured. It's shorter than the time taken to manually label multiview 2D keypoints, but it's definitely (much) longer than the record time. Most of the data cleaning budget is spent on complex tasks in OakInk2.
5. Motive 3.0 Body or Edit. Also, there is custom-developed software based on Qt and Blender to meet the specific needs during dataset construction.

Constructing and curating datasets is very labor-intensive work. I hope these answers can help you!

[ryhara]

@kelvin34501

Thank you so much for your detailed responses!!
The information you provided is very helpful and informative.

I’ll reach out again if I have any further questions.
Thank you again for your support.