simtk.org

I previously collected data for an ergonomics analysis on August 8th and 9th for my dissertation research. However, I got very large and unrealistic numbers for RMSE and MAE for majority of the upper body. I saw previously that this was a known issue in the Forms, and OpenCap has recently released a new model for upper extremity analysis.

I was wondering if someone would be able to assist me, I tried to run it locally but on my personal computer it is take 6+ hours to run 1-60 second video. I have also tried to utilize my universities HPRC, however I am unable to install OpenPose and recompile as a Linux file.

Does anyone have any advice, the following sessions are what I need reprocessed.


session_ids = ['3b3db40f-d74b-4b1d-aa37-45b1e582787a', '842a86b2-c744-4537-a626-15f51671268f', 'd7023773-b1ec-4a76-8df9-2651377cfa2e', '0339b8d0-f89c-4a75-b1c6-09234cf680cb',
'0192c37e-ffc4-4c18-82e9-6b37ce9e1319', 'bbeac9bb-a3fb-475c-9b30-fce780750a5b', 'f71101aa-a0cf-4195-a179-97d3a2801a6f', '487ce472-db6c-4f3e-adc2-fb0233ef39b5',
'3509ffb5-3280-40d6-a56a-e530b95baae0', '236d68e5-8ce8-4e17-86cc-6620df66cc22', 'eb723f44-8ea9-4445-8031-aa222d619254', '3e36d012-7c67-41f3-867c-394a96f578e1',
'e044fb57-9876-4431-87d4-7096865d5bd5', 'd024eff9-93b8-4157-81a3-05fa4195f998', '13d9dee8-f58f-49c6-a2df-14f157be855a', 'ece57235-51c0-4b27-8aa6-0e5b8ca94f6d',
'974736a1-4b4c-444d-ae34-dcefc9033394', '35fb9afe-68ce-4e72-957d-5e7807df5bcd', 'e8614a1e-5eaf-418c-8fda-43ad6f99061a', 'fd3acaa8-4efd-4573-ae89-87915c045e0d',
'b9462048-7a12-4456-908b-00d6c5c8c7ca', '726f0ed5-7b2b-4532-a968-7f006aca2ff3'
]

Any help for a poor PhD candidate would be appreciated... I'm so close

Hi Whitney,

The new model will increase the DOFs at the shoulder (6 DOF instead of 3 DOF). Do you think that will improve the motion you are interested analyzing?

A few other notes looking at these trials. For long durations the hands are occluded which makes it hard for the pose estimation software to give you accurate hand/arm kinematics (which may not be corrected by reprocessing). Also, when the participant is wearing dark clothing that matches the background during smaller movements like this, it makes it hard for the pose estimation to see joint centers (which it is probably why some of these are failing).

Thanks for the response. We thought this might be an issue, and will be great in my limitations of my study.

I am not sure if 6DOF will improve the errors however we wanted to rerun just to make sure. The errors I currently have for the arm and elbow for RMSE and MAE are all greater than 100. However, if this is expected with occlusion and my current environmental factors than perhaps rerunning would not yield any better results.

Hi Whitney,

To reprocess your data with the other model, please take a look at the video in the post in the OpenCap news (https://www.opencap.ai/news) named 06/09/23 - Musculoskeletal model with ISB shoulder model. Let us know for questions,

Antoine

Hi Whitney,

As Antoine mentioned, you can use that code to reprocess. I'd suggest testing it one of your sessions and see if you see an improvement in kinematics. If your MAE is that large though, it's most likely you won't see much of an improvement, but you can check.

Hope this helps,
Matt