Spine still bends oddly

Greenlaw · **Posted:** Sun Jan 27, 2013 6:57 pm

AaronC wrote:

The bent spine issue has come up for me also, and I'm experimenting with reducing character height and finding some success.

Yes, it does help (as described above,) but I have to scale the character considerably shorter to get the spine and hips to stay somewhat aligned when crouching, and this causes problems elsewhere in tracking because now the rig is too short to deform properly.

FYI, this was not a problem back in 1.0, not even in its early days--for example, here's a test I did over two years ago using 1.0 with similar motions (crouching, stretching, etc.):

iPi DMC 1.0 Deform Test

If you look closely, you'll see that the squatting motion does not have spine bending or hip problems at all. For that matter, the shoulders stay symmetrical too. The tracker and rig in 1.0 did not have the slightest problem matching these poses.

Another example is Happy Box, which was almost entirely shot in sitting positions with 1.0, and the spine/hip problems did not occur there either. Here's are two examples of iPi DMC 1.0 data that was created for the film:

Dual Kinect test, iPi DMC 1.0

Single Kinect test, iPi DMC 1.0

One more point worth mentioning: back in 1.0 days, I typically set the rigs to be slightly taller than my actual height, which was also the recommendation of the developers.

G.

Greenlaw · **Posted:** Mon Jan 28, 2013 7:12 pm

I revisited Studio today to see if there wasn't something I missed. Unfortunately, I still could not prevent the Right Shoulder and Spine/Hips errors.

However, I did find another way to repeatedly demonstrate the problem with the right shoulder:

Greenlaw · **Posted:** Fri Feb 01, 2013 10:07 am

Hi vmaslov

It's been a week so I was wondering if any progress was made. I haven't heard anything from Michael or the other programmers but I imagine they are extremely busy right now.

My top question: were you guys able to reproduce the errors with the files I sent? If not, this might tell me there is a problem unique to my configuration because the errors are present every time. I can submit more examples if it helps.

On the other hand, if you are able to reproduce the errors, I have additional information I discovered a few days ago.

First, set this up: Open and 'quick track' the Shoulder Test scene I submitted, with Shoulder Tracking enabled. Now go to frame 217. What you should see is the actor with both arms fully raised above his head, the left shoulder in extreme high position (correct) and right shoulder in flat level position (incorrect.)

Now, if I set Trajectory Filter to 0 and click Refit Pose, the right shoulder will correct itself into the high position. However, when I click Track Backward or Track Forward, Trajectory Filter will set itself back to 1 and the shoulder will pop down, returning to the opposing shoulder rotation problem.

Another important observation: I can only get Refit Pose to correctly track the shoulder for the current frame if I first deactivate Enable Shoulder Tracking, click Refit Pose, reactivate Enable Shoulder Tracking and then click Refit Pose again.

Probably less important but worth mentioning: If I set Trajectory Filter higher than 1, like 3 or 4, and click Track Forward or Track Backward, the setting will persist. However, if I set to 0, it will always reset to 1 after using either tracking button.

In other words, iPi Mocap Studio will not allow the persistence of Trajectory Filter level zero even though it seems to require it for correctly tracking the right shoulder.

I don't know if the above has anything to do with the asymmetrical spine and hips problem but I suspect there is a connection.

I hope this information is helpful.

If you can, please keep me informed on your progress. We have a lot of other work here to keep us busy in the meantime but we hope we get back to tracking with iPi Mocap Studio soon.

Good luck guys! :)

G.

Michael Nikonov · **Joined:** Tue Mar 31, 2009 9:13 pm **Posts:** 805

We still research the problem.
Shoulder problem is unlikely to be something that can be fixed quickly.

Greenlaw · **Posted:** Fri Feb 01, 2013 10:46 am

Thank you for the response Michael! I'm glad to hear you guys are still on it and I'm confident you will squash this one. :)

Okay, that's actually good to know. Rather than wait, we will think of ways to work around these issues until there is a fix then.

G.

Greenlaw · **Posted:** Fri Feb 01, 2013 11:12 am

Right now I'm looking through our raw untracked motion data for the second Brudders movie. I have to admit, while the motion in this movie is more complicated than what we did for Happy Box, it's probably not going to be too demanding of iPi Mocap Studio and we should be able to deal with this problem without a lot of extra work.

That said, the character performance we have planned for our next movie is going to be much more sophisticated and we're definitely going to need the shoulder/spine fix by then--but, seriously, we're months away from starting that one.

So for now, all is good. :)

Greenlaw · **Posted:** Fri Feb 01, 2013 11:37 am

I just noticed something very strange. This is definitely a bug but I think it's a good one--at least for now anyway.

I opened a clean, untracked project for the Shoulder Test video but this time I left Enable Shoulder Tracking disabled from start. Forward Tracking was moving along quite speedily, and when it reached the point where both arms are fully raised, this time the tracker correctly tracked both shoulders!

So, important tip for other users: if you want shoulder tracking on, leave Enable Shoulder Tracking off. At least for now. (Note: The current build as of this post is 2.2.1.145)

Here's a screen cap of frame 217, arms at max height. There is still something funny going on with the right arm--we'll just have to deal with that--but the right shoulder does look good now.

Attachment:

2013-02-01_103422.jpg [ 38.7 KiB | Viewed 16621 times ]

G.

Greenlaw · **Posted:** Fri Feb 01, 2013 11:49 am

Small update--shoulders correctly move up but I'm not sure they come all the way down when arms are at lowest position. However, they do stay symmetrical when both arms are down so this is still preferable to what happens when Shoulder Tracking is enabled.

I think this shoulder positions are fine right now but in a future build, it would be handy to have sliders that sets the min and max rotations for shoulders. Thanks for listening. :)

G.

Greenlaw · **Posted:** Fri Feb 01, 2013 11:59 am

Now this is very interesting:

I was switching between Low Resolution Tracking and High Resolution Tracking, and sometimes I got more accurate results with Low Resolution Tracking. The trick seems to be to use Low Resolution for frame ranges where trouble begins to occur and then switch back to High when the trouble spot is passed.

Or could it be possible that the options are working in reverse? I need to test for that.

The result in the image below does look much better. To achieve this, I used Low Resolution Tracking while the arms were being raised and switched back to High Resolution after they arrived at the peak height.

Attachment:

2013-02-01_105457.jpg [ 39.32 KiB | Viewed 16620 times ]

I'm still investigating 'workaround' solutions but this one seems most promising.

G.

Greenlaw · **Posted:** Fri Feb 01, 2013 1:01 pm

After working a little more with the Low Res/High Res tracking trick, the right arm is still not quite right but it is better than it was. Just wanted to let users know that they shouldn't expect this to be a perfect solution.

Disabling Shoulder Tracking does appear to result in better shoulder tracking though. Note that you should do this with a clean project--if you have already used Shoulder Tracking in your project and then disabled this feature, the result may be uneven.

G

Spine still bends oddly

Who is online