Actor Scaling Bug

Greenlaw · **Posted:** Sun Mar 25, 2012 11:34 am

Hi,

I'm finally getting around to testing iPi DMC using a little performer.

You might recall from over a year ago when my then four year old daughter and I did a test and we discovered that iPi was not capable of tracking 'child' size performers. You guys fixed this promptly and I really appreciate that. My apologies for not being as prompt in testing this 'new' feature but my now five year old daughter is still very 'child sized', so here we go. :)

Anyway, in doing this test I think I discovered a bug in the latest release (127): When scaling the iPi Actor, the Actor is scaling from it's center instead of it's base. This is probably a new error as I don't recall seeing this issue with previous versions. (Or I may not have noticed since I typically load a custom Actor for my own body proportions.)

If possible, can this issue be fixed soon? While it is possible to work around this error, it does make setting up a T-Pose more difficult because the Actor must also be positioned along the Y-axis. This is made even more problematic when the Actor needs to be scaled down to just over 1m. If this error can't be fixed soon, can you post a link to a previous release where this was not a problem?

Thanks in advance for your help! If you need more information, please let me know.

G.

Greenlaw · **Posted:** Sun Mar 25, 2012 11:56 am

Ugh! This error is worse than I first thought. After scaling the actor I cannot reposition it along the Y-axis to align the feet to the ground plane. I'm guessing that iPi Studio thinks the ground plane is wherever the feet are after scaling.

This is probably the same error some users complained about recently where the actor appears to 'float' above the floor. As described in the previous post, the error happens because Actor Height is scaling from the rig center instead of its base--in other words, the feet will rise above the floor when the actor is scaled down. Unfortunately, the error also prevents the user from correcting the actor's position relative to the ground plane.

If you need more info, please feel free to ask questions. If you have any suggestions, please post. In the meantime I will have to set iPi DMC aside until this issue is fixed.

Thanks in advance.

G.

Greenlaw · **Posted:** Sun Mar 25, 2012 12:00 pm

One more observation:

If you scale the actor up, it causes the feet to penetrate the floor of course, but iPi Studio does allow the user to move the actor 'up' to align with the ground plane. If you scale the actor down, causing the feet to float above the floor, iPi Studio curiously does not allow the user to move the actor 'down' to align with the ground plane.

As mentioned above, this is in version 127.

G.

Greenlaw · **Posted:** Sun Mar 25, 2012 3:29 pm

Another curious discovery: if I click Pose Refit, the rig flies away--that's not surprising because I am not able to position the rig anywhere near the actual performer. However, I found that doing this 'releases' the rig so I may now position it to align correctly with the ground plane. Of course, now I need to individually reposition all the limbs too, so this not an ideal situation.

I'll post more info if I learn anything new and actually get this working.

G.

Greenlaw · **Posted:** Sun Mar 25, 2012 3:40 pm

New discovery: I tried regressing to build 126 and the bug is present. Next I tried regressing way back to build 119, and the bug is present there too.

I guess it's been around for awhile but few users noticed it because not many have tried scaling the iPi Actor down to 'child' size, or maybe because they always use pre-made Actor files like I do.

However, I think this explains the issue some users have reported about 'floating feet'--if you scale Actor down smaller than the default size, the iPi rig does 'float' and you cannot translate it down unless you do the previously described Pose Refit trick to 'release' the rig.

Hope somebody will find this info helpful for the current release, and hopefully fixes will come soon.

G.

Greenlaw · **Posted:** Sun Mar 25, 2012 3:43 pm

Downloading the installer for 127 now...since the bug exists in older versions too, I may as well use the latest. :p

Greenlaw · **Posted:** Sun Mar 25, 2012 3:58 pm

Wow, my daughter is tiny and skinny (1.07 m). Of course the problem last year was that she was a bit shorter than 1 m. At least she's growing. :p

Greenlaw · **Posted:** Sun Mar 25, 2012 6:04 pm

I wonder if iPi DMC 2.0's additional Actor models will include one optimized for tracking a small child.

I'm having some issues tracking this data. It's partly because I'm using the 180 degree setup which I've found to be far less precise than the 90 degree setup, which I'm guessing is because the mesh has greater density when you using 90 degrees due to the overlapping data. The other reason is because five year old children move like total spazzes and it's freaking the software out. :p

Also, I think I'm seeing occasional volume mismatches between the two data streams. It could be due to the subject's fast movements and also because the two Kinects are about 16.7 ft apart from each other and I think this distance may be pushing the range of the devices. The mismatches look like what I described in another post, kind of like a temporal offset. It's puzzling because the data seems to go back in synch. I'll need to check later to be sure but I don't believe there are any dropped frames in this data.

Michael or Andrew, would you like a copy of this data? It may present some interesting ideas to think about.

G.

Greenlaw · **Posted:** Sun Mar 25, 2012 7:05 pm

Another update: I regressed way, way back to build 112 (just because I can,) and I see that the scaling from center issue exists there too. I guess this means this is not officially a bug.

That said, I think the rig probably should scale from the base instead of the center of the rig (the hips?)

Still not sure why iPi Studio doesn't want me to translate the rig down to align with the ground after scaling it smaller. This behavior, I think, probably is a bug.

G.

Greenlaw · **Posted:** Mon Mar 26, 2012 10:39 pm

I thought about the tracking errors I got yesterday and arrived at two new conclusions:

1. The temporal error we've been seeing is probably the Kinect equivalent of a motion blur artifact. I think 30 fps is simply not sufficient for capturing/tracking the kind of motions my five-year-old daughter was making. My feeling is that in this situation it's probably best to use the PS3 Eye system with it's 60 fps frame rate. Next week I will try to talk her into another shoot with the 90 degree setup and the six PS3 Eye camera setup and compare results. But before I do that, I will try a controlled series of repeatable motions myself and compare results. Will let you know how this goes.

2. The iPi Actor in version 1.0 is not optimal for tracking children, especially small 'skinny' ones. I noticed that even with the Actor torso and limbs scaled as thin as possible, it does not fit into my daughter's volume space correctly. I am wondering, will the upcoming iPi DMC 2.0 include an Actor optimized for tracking children?

Thanks in advance for any input. If you need more information or data from me, please let me know.

G.

Actor Scaling Bug

Who is online