Marker positions in 3D or 2D about arctic HOT 7 CLOSED

Hi Amin, thanks for your interest.

I will work on releasing the mocap data this month. I will let you know in this issue once it is ready.

Alex

from arctic.

@heydarshahi

MoCap can be downloaded and visualized now. See changes in this commit.

from arctic.

Yes it is in mm (https://github.com/zc-alexfan/arctic/blob/master/docs/data/data_doc.md) as mentioned in the docs.

The mocap data is the raw data, which was in mm and then in my postprocessing I used meter.

from arctic.

Thanks, Alex! Looking forward to the release of the marker data.

In the meantime, I wonder if the joints_gtvariable used in eval_modules is actually coming from the marker positions?

If that's the case, looking through the code, it seems to me that ground-truths are being computed, but not stored. Is that correct?

If that's also the case, is this data coming from splits? Or is it coming from the leaderboard data?

from arctic.

joints_gt is the ground-truth 3D keypoint of the MANO hand. So it is not directly related to the marker positions. However, MANO was registered to the markers during MoSH++.

from arctic.

That's amazing, thanks!

from arctic.

Re-opening for a potential discrepancy between the released data and previous data:

The previously released data regarding 3D points (e.g. egocam markers, kp3d, etc.) have a small range as they are supposedly measured in meters, while the newly released mocap data seem to be measured in millimeters as they are in a x1000 higher range. I'd suggest to either have the mocap data in meters to keep consistency with the previous data, or to add a line to the documentation explaining this to avoid potential confusion.

from arctic.