Build 114

Greenlaw · **Posted:** Sun Mar 20, 2011 2:36 pm

After a long absence, I finally made time to work iPi DMC again, and I'm using the latest build 114 for the first time. So far I think this is an excellent update! The tracking results are superior to what I was seeing in previous builds. More importantly, I'm not seeing any of the errors I reported from the last build. Also, 'pre-jitter removal' jitter seems greatly reduced, and tracking speed is very nice too. (I'm using a GTX 460 and iPi Studio reports that I'm averaging about 0.67 per frame.)

Needless to say, I'm very pleased with this build. Thank you iPi Software! :)

I haven't tried the new Recorder with the six PS3 cameras yet, but will do so soon. I will also do some more tests using Kinect, maybe tonight.

Regarding my short film project, we solved several import rigging problems the other night, and the mocap data is coming into our animation program perfectly now. Finally, I feel we can move ahead and finish this thing!

BTW, if you guys can introduce the new head tracking feature any time soon, I would love to test it. We still have a lot of mocap to shoot and edit and I can still use this feature immediately.

Keep up the great work guys!

Greenlaw

P.S., Sorry for the long delay in getting you a high-quality version of the Deform Test video. I'm working on that now and will post it for you tonight. Also, I will try to create a similar six-camera version of the test video later this week and give you the file if you wish.

Greenlaw · **Posted:** Sun Mar 20, 2011 2:55 pm

Examining more closely my 2500 frame track of the Deform Test using build 114. I really only see three joint pops that need to be addressed, and I feel two of them probably would go away when this video is reshot using the six camera setup instead of the four. The third pop occurs during the arms folding, which wasn't there in the original track from a couple of months ago, but I understand that this is a hard motion to track and it really wasn't perfect in the original track anyway. In any case, manually fixing these three pops should be trivial in iPi Studio or most other 3D applications.

I still see an issue with the feet during the twist, but I expected it and I understand that this problem is also difficult to track with the current rig. The problem is easy enough to fix in Motion Builder (or even inside of Lightwave) though so I'm not concerned.

Great job guys! This version seems to be a big improvement and I'm looking foward to the next version with much excitement!

G.

Edit 1: Simply tracking backwards though the pops seems have fixed them. Can't get simpler than that. Thumbs up! :)

Edit 2: The arms folding, on closer examination it's actually much better than I first thought, definitely better than my original track even.

Greenlaw · **Posted:** Sun Mar 20, 2011 8:31 pm

Michael,

I will email you the link for the Xvid version of the Deform Test video in a few minutes. From what I can tell, the properties of this file appear to match the original 1280 x 960 59fps RAW file. Let me know if this is suitable for your use; if not, I can try again.

For everybody else, I will make this file or a very similar one publicly available at some point in the future. FYI, I've found this file to be very useful for testing rigs and deformations not only in iPi DMC but also in other 3D apps.

BTW, I'll be posting a new video to the LGD Test Tube channel showing the difference Build 114 makes, following through with an example of the new Deform Test motion applied on the Sister character. Watch for the link soon. :)

G.

Greenlaw · **Posted:** Mon Mar 21, 2011 12:01 am

Right now I'm using Build 114 to work my way through re-tracking all of the 'Pooper' mocap data, and so far the results have been much cleaner and more accurate in every shot!

G.

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

In build 114 we modified arms tracking algorithm. It is still not perfect but it covers bigger range of motions then build 113. So you should see some difference in arms tracking. We did not change leg tracking algorithm so there should not be difference in leg tracking compared to build 113. If you see any difference in leg tracking it should be attributed to the stochastic nature of optimization algorithms we use. Basically, results may vary a bit from one tracking run to another, this is how current generation of our technology works.

Tracking backwards should resolve many tracking errors. Since we use (relatively) small number of cameras, the system has to fill in missing optical information based on some other factors (continuity constraints, kinematic constraints etc.). There is a certain contradiction between optical constraints (what computer can see in current frame) and continuity constraints (how the pose in current frame corresponds to poses in nearby frames). It is not always easy to find a good balance between these two factors. If we give more weight to optical constraints this may lead to speculation based on optical information being too aggressive, and there will be discontinuities in animation (like unnatural sudden twist of arms etc.) From the other side, if we give bigger weight to continuity constraints then this may affect the system's ability to track fast motions like dancing. Ideally, we should process the entire video clip (or at least do a look-ahead of a few seconds deep) and only then reconstruct the animation corresponding to the video, but that would be too time consuming and impractical at the current level of technology. So the current algorithm is designed to work based on forward tracking (without look-ahead), and user has an option of activating backward tracking manually when forward tracking is not enough.

Greenlaw · **Posted:** Mon Mar 21, 2011 11:29 am

Yes, the most noticeable improvement I see is in the arms. My characters are playing musical instruments and previous version of iPi DMC could not pick up on small and subtle arms/hand movementa. Build 114 is preserving much of the small motion quite well. More importantly, I'm not seeing the frequent pops I was getting with 113.

Overall, 114 is giving me the best results ever!

I did run into two issues last night but they were minor and easily corrected. Unfortunately, the errors seemed to occur randomly:

1. Sometimes when I use Jitter Removal using a different setting over a short range of frames, I may see a small pop where the transition occurs. I know I can correct this pop by moving the Trajectory Filter slider to 0 and then back to the desired value, but this wasn't always working. That said, if I re-do the Jitter Removal over a wider range of frames, and then do the Trajectory Filter slider trick, the pop usually goes away. The impression I get is that sometimes Trajectory Filter may not be smoothing the curves as expected, or maybe it doesn't work predictably if the selected range isn't long enough to cover a complete motion.

I think this happened two or three times last night. I don't know how to predictably repeat it but 'll try to catch a repeatable example to show you.

2. In one instance, after I ran Jitter Removal with a different setting over a short range, Jitter Removal for the rest of the footage was mysteriously removed. This glitch seems more serious but I haven't been able to reproduce it, so maybe it was just a fluke. I will let you know if it happens again.

I hope to be working with iPi Recorder and DMC every night this week, and I'll let you know how it goes.

Very exciting time guys! :)

G.

TheOwl · **Joined:** Mon Oct 12, 2009 4:39 am **Posts:** 84

I also want to verify build 114's MAJOR improvements.

Before I was only using low resolution tracking because it runs on 1.9 secs per frame vs high res which was 3 secs per frame.

With the new build it is now down to 1.9 secs per frame on a high res setting!

And the tracking is very accurate now for me. Before I had to repeat the tracking with alot of troubleshooting. Now all I need is just one tracking run and its all I need!

Thanks to our Russian friends in ipiSoft!

Build 114

Who is online