Is there any point to trying to squeeze out more performance (better/multiple cards, etc), or are tracking speed improvements to the software going to make this wasted effort/money, soon? My specific concern being the next 60 - 75 days. Mostly, I think I'm fine, with what I have, but ... there may be cases late in the project where quick turnaround will be very helpful, so ... thought I'd ask.

Thanks.

What kind of performance are you currently getting? (Seconds per frame.) What's your target speed?

I'm currently getting a little over 0.5 spf in Mocap Studio 3 when tracking data from 2 to 3 Kinect sensors, and I'm using a fairly ancient GTX 460. That's slower than I like but it's been fast enough for most of my recent needs.

I'm planning on upgrading my gear later this year though. My current computer is starting to feel old all over. (Or maybe it's just me.) :)

G.

_________________
Greenlaw
Artist/Partner - Little Green Dog | My Demo Reels (2013,) (2015,) (2017,) and (2019)

Watch a one minute excerpt on Vimeo now!

Culled from my post in General (22.5% throughput improvement?):

With 6 cams, and everything checked, plus shoulders from Video, and flexible spine
And the gtx960, I just added, I get 1.15 spf.
If I have 2 captures to process, I can open up 2 copies of Studio, and get just under 1 spf, cumulative.
--------------------------------------------------------------------
By turning off head tracking, and shoulder from Video, I can speed it up considerable, but these features have been a major improvement for me, particularly for motions that are crouched over.
Of course, I took a hit, moving from 4 to 6 cameras as well.
The gxt460 was managing about 2.4 spf. A pair of 460's ... about 1.5 spf, and a 750ti, about 1.38 spf (1.12 with 2 Studios). The 750ti, considering it ran on a single 6 pin power connector, was pretty impressive, I thought, though I should point out, it was the top of the line evga ftw. BTW, the 2 Studio trick worked for the 750ti, and 960, but not for the 460 (never tried it with a pair of 460's).
Wish I'd stuck with the 750ti ... and added a 2nd ... though with the 960, I could always add another 960, if I see a sale and some rebates at some point.

Target speed? Fast as I can get ... what I have is fast enough for now, but I expect there will be moments where I'll be in a crunch to plow thru some replacement motions in a hurry. Faster than realtime would be best ... then I could finish before I started :)

Mike

Okay, thanks for answering. I was asking mostly out of curiosity and for comparison. I think my time is shorter because I don't use head tracking (I rely on the Move controllers for that,) I sometimes disable shoulder as well but that usually depends on the results. Also, it's probably not fair to compare times between Kinect and PS Eye data since the tracking data is differently. (I suspect multiple Kinect 1 data is smaller than multiple PS3 Eye data, and multiple Kinect 2 data is probably a lot larger than multiple Kinect 1 data.)

I have noticed an overall improvement in tracking speed between iPi Mocap Studio version 2 and version 3 though. No actual benchmarks, just an observation.

G.

_________________
Greenlaw
Artist/Partner - Little Green Dog | My Demo Reels (2013,) (2015,) (2017,) and (2019)

Watch a one minute excerpt on Vimeo now!

Regarding head tracking.

The effect on the rest of body, from using head tracking was beneficial enough to make it indispensable, for some of my motions. My understanding, is that although using a motion controller for the head tracking removes the hit on processing speed, it would give me head orientation (which I do need), but would not give me the impact on the rest of the body, that has proven so useful. I do have a head harness, and will be capturing with motion controllers, to verify this, one way or the other, when I do my next set of captures. My conclusion was based on a single post I read, and ... easy enough to test.

And, yes ... comparing various #s of Kinects, vs various #s of PS Eyes, would seem ... apples/oranges, though it sounds like the processing time would be in the same general ballpark, after you eliminate as many differences as possible.

Mike