Hey there!

I'm quite new to iPi Soft, but have been studying it thoroughly and experimenting a lot. So first, thanks for all the knowledge shared around here!

Using a 6 PS Eyes set-up, I have already been able to pull out of it some pretty decent data.

However, I have a session - already captured - for which the same calibration file is failing to be optimal for all takes. We only had time to do a calibration at the beginning of the session and another at the end, and they result a little differently. Our mounts were probably not sturdy enough, but, that said, I'm trying to get the best I can out of what I already have.

I've managed to get better tracking results by manually moving the camera after calibration. I know it is not the recommended procedure, but it is helping. However, when I move any camera, the "roll" value is instantly zeroed out. Do we have any access to be able to modify this value or not?

Not everything always goes the way we planned according to plan, and if the software offers broader access to its setting, it can be a real life-saver!

Hi
Indeed, there is no control for changing the roll angle of the camera. Mainly because the purpose of editing controls is to set estimated camera positions before calibration rather than to adjust afterwards. However, you can edit the iPiScene file manually using any text editor in case you know what position and rotation you need.
There is also a chance that camera were not moved during recording, but both of your calibrations produce inaccurate results due to some factor, say poor environment conditions for calibration. You may share you calibration video so I can check this guess.

Thanks a lot for the reply! I was naive not to try to open the parameters file on a text editor. It's good to know.

Anyway, if you have the time, I am sharing the two files. It's quite big, as I realize the background for the first session doesn't seem to be properly measured. You might be able to better evaluate this.

https://drive.google.com/file/d/0B1v2rH ... sp=sharing

Any further insight is appreciated! We are still getting used to the software and its best practices.

...

In my experience with a 6 cam set up, I personally chose to use a cam 1 height of 1.50m and it really doesn't matter as much that every camera is exactly what the actual cam height is after I set that camera at that level, though they will be close, they can be off up to 3-5 cm sometimes and it really doesn't affect the tracking, but the actual actor height should be very close to the actual performers height +/- 1 cm at most.

I am not sure how you do your calibration, probably how the iPi Docs show, but I found it very critical to have an exact center point marked on the floor and to have that mark appear very close to the same position in every cam view port pertaining to the floor positioning on screen, pointing at chest height, not at the floor mark, this has nothing to do with where you start your performance, or aligning the actor for tracking, it is for calibration accuracy mostly, but it can be used as a start/finish point also.

I did have an issue with this recent update where the scene calibration file somehow changed on its own and was severely off, after I had tracked several videos without issue prior, so I had to re-run the calibration video and all was fine after, that has never happened before, so it threw me.

It will take you a bit to work through the process on your own, but the program does work very well for me anyway with a 6 cam set up, but I do use a different set up and calibration technique than is shown in the Docs, due to my camera positioning, I can't use a high sweeping light track cage, but I get excellent calibration.

Trying to over think the process will probably cause you more issues than just letting the program adjust its self to the camera heights in my opinion.

We all had issues when first started with it, but if you stick with it and work through things it does a very good job, consistently.

...

Thanks for the tips, Snapz (and also for all insights shared around here; really useful stuff!)

I am indeed in the first steps of getting a consistent workflow out of it, and I only had previous experience with inertial systems, not optical ones. So it's a matter of further exploration

I got curious about the data quality some people are getting from it (you included). I have yet to find a setup that allows me to use the actor height that close to the performer's. So far it is not consistent at all (one case showed a 15cm difference). Camera height have been consistent, what surprised me was some apparent shifts on its place horizontally. That was why I placed this question, so I could save some interesting data I wouldn't be able to re-capture.

I'm still struggling with the effect of some settings. Lighting, for instance, proved to cause dramatic changes. The different options for shoulders too. I keep head tracking on, as it's leading to better spine tracking, but I'm sticking to the stiff spine setting, for the flex one - which I believe you use - caused some really twisted bad stuff.

If people like you hadn't shared positive experiences using such settings, I'd probably consider they were less appropriate choices, instead of going back some steps and trying to get a better set-up and calibration. For that I thank you.

...

Well, sadly once you record a session with a calibration, you can not just re-calibrate and have it work again once the cameras have been moved, so until you get a consistent process of calibrating before you record any final performances, you just have to write those others off as tests.

I personally also use a very rigid fixed camera mount to the walls, so really my cameras never move now, but that isn't necessary, tripods will work, you just have to be more careful around them, as even the slightest bump will change the cams calibration, why it is best to do 2 cals, one before and one after, always.

My calibration consistently is .11 - .15% misdetects and a reprojection of less than 1, (usually less than .65 on all cameras), so if it falls off that on any calibration attempt, I know something didn't go optimally and I re-do the calibration, but that is rare now.

Very bright room lighting will affect the calibration, even when extra darkening mode is used, best to have mild ambient lighting while calibrating and use darkening mode, do not turn off all lights though, it won't be accurate in my experience.

With an accurate calibration, the actor should match the performer very closely, and should basically for the most part remain overlaid throughout the performance, but the closer you get to one camera you move farther away from others, so a slight height appearance will occur, but it's not to worry about, it is normal and shouldn't affect the tracking.

What is your capture volume by the way? X m by X m or (ft x ft)

As far as the flexible spine, you should never use stiff spine, unless you need no spine action, and it will also throw off the tracking for a smooth human animation.

I use the very flex spine for the most part, and in some areas I switch back to flexible when the action is too extreme and it throws the hips too far forward especially on extreme dance moves with bending at the hips and the body rolling back up to vertical, you will just have to get the feel for how to use the different settings and when.

I can only speak from my experiences and it probably took me the better part of 3 months to be able to get the process to work as I wanted, because I use hired dancers and I can't have failed takes to just have re-takes at a later date, so a lot of trial and error and camera re-positioning was needed, but I think I have it working as good as it will now and I don't have to really limit the performers actions, except to not hide hands or feet from the cameras and always keep about a 6 inch perimeter off any body part when possible, (body, legs, head), the tracking works better and just easier to work with the animation in post that way.

I feel I get fairly life-like animations, with very little clean up, so I am happy with my outcomes.

The program with 6 PS Eyes works well, so it's just working through the calibration, which is the most important process, other than using tighter style clothing in darker saturated colors works best, and darker color gloves helps too.

Really you should not have to do very much, the program is pretty automated, except for the actor parameter choices and if you do use jitter removal, don't set any body part above "1", except the head only, you may want to add more to stabilize it better.

When calibrated correctly, your actor height should match your performer height closely, so to me it sounds like the calibrations cam off floor height for cam 1 is off, for the rest of the cams let the program set the height and position, even if not exactly the actual heights, within reason of course, (+/- 1 to 5 cm is fine), as they are just numbers estimated by the program, but the actor in studio should remain the appropriate height and positioning in all cameras throughout tracking.

Good Luck!

...

I like to record calibration data before I start a motion session and immediately afterwards. This is valuable insurance that costs only a few seconds of extra work. If I'm planning to record a longer series of motions, I may even record one or two more calibrations in-between scenes because, the way I see it, the longer I'm recording data, the greater the chance that a device/camera may get bumped or otherwise moved just enough to ruin the entire session, and the extra calibration data recordings will help cover that.

When I'm not in a terrible hurry, I'll even take time to process the calibration in Mocap Studio before I start capturing motions. I never assume everything is going fine and I like to spot check my work during the sessions to be sure. To me, that's a few extra minutes well spent because it's a terrible feeling to have to scrap a couple of hours of work because of a botched calibration video.

Normally, I never need to use the extra data but I'm always grateful to have it for that rare occasion when something does go wrong.

_________________
Greenlaw
Artist/Partner - Little Green Dog | Demo Reel (2017) | Demo Reel (2015) | Demo Reel (2013)


Watch a one minute excerpt on Vimeo now!

I've looked at the files. The results of calibration look reasonable in both cases: detected positions of cameras correspond to their images in video. However, there are noticeable changes in positions of cameras #2 and #6 between 2 calibrations. I guess these 2 cameras were moved from time to time during the session, thus there are action videos with camera positions not matching any of the calibration videos.

I suggest you to try tracking without badly positioned cameras. You can cut out unwanted cameras from iPiVideo using the edit feature in iPi Recorder.
http://docs.ipisoft.com/iPi_Recorder_User%27s_Guide#Editing_and_Export
To remove cameras from iPiScene file, remove corresponding tags using a text editor.

One more note. In the first video I see quite a big grain in image from camera #1. This is not good for tracking. Either this camera is deffective or it needs brighter lighting for solid picture.

...

Yes, you can edit the video in iPi Recorder to remove bad cameras, (this may really affect your tracking errors though, depending on which cam was faulty), but to clarify what Moslov stated.... you MUST remove those same cameras in the Scene file also in a text editor and you can re-save it as a different name to keep the original, if this step isn't done the file won't open in iPi Studio, it will throw an error stating the number of cameras don't match.

True, the room lighting is better when equally ambient, not uni-directional or extremely bright while recording, this will cause too much variance in colors on the iPi actor mesh and confuse the tracker resulting in more tracking errors to deal with during the process.

You can also play with the gain and exposure settings for each camera in the set up process of iPi Recorder, not all cameras have to have the exact same settings, but also shouldn't have drastically different settings, this should be done prior to any recording session anyway to get all cameras looking good in the view ports first. (I also always turn off the auto-gain control).

If you move a camera on purpose, or inadvertently, or it drifts without your knowledge during a recording session, you MUST re-calibrate the cameras in order to keep all cameras usable, of course if it drifts without knowledge, you are kind of stuck and would have to remove the bad cam later as stated, or re-calibrate at regular intervals during a session, which to me would be kind of a pain and easily avoidable.

Not all bases on every PS Eye will be tight, especially if bought used, and just the weight of the cable can cause drift, so taping the cameras to its base, or at least to tape the cable to minimize pull on the camera is a good practice.

Most of these issues are why a lot of people opt to use the Kinect sensors with iPi for easier set up and then just clean up the motions in other editors, but they too have their own limitations and dual, or multi-XBox 1 cameras each require a separate computer and distributed recording to work, (as of now), as well as both computers even being able to run the XBox 1 sensor, not just any computer will run an XBox 1 sensor.

You can still run dual or more XBox 360 cams on one computer, (and more compatible with more computers), with just about the same results as the XBox 1, you will just have a lessened FOV and video quality, the tracking quality is almost identical to me, but I don't use sensors on a regular basis to really speak too much to what settings, or adjustments can be used to improve the tracking quality inside iPi Studio, I just don't need both set ups in one studio, when my PS Eyes work fine, night or day.

I will say that the Kinect sensors are a more portable set up for sure, if you wanted to record outside of a single studio environment and not have to worry as much with lighting conditions, (always avoiding sunlight), the clothing constraints, or harder calibration issues, just really depends on individual user expectations of output.

...

I can only thank you gentlemen for the feedback! Really insightful, and this kind of support is what enables us new users to achieve a steeper progress on our learning curves.

I have overlooked the option to exclude data from one camera before bringing it to iPi Studio; it might be very helpful.

Sorry for the late reply, but the last setup I had had an active area of about 5m x 3m (it's an asymmetrical shape, but including cameras would take around 8m x 5m).

I'll have a new recording session soon and I'll surely be more attentive to the details you all have pointed out.

Again, thanks a lot!