Calibration error

[lambdaloop]

Calibration error

[lambdaloop]

Hi @menor22,

That's an interesting error. It seems that I had the wrong threshold for the number of points needed for pose estimation of the checkerboard. It should be 4 instead of 3. I fixed this and updated calligator.

Could you try updating calligator ( pip install --upgrade calligator ) and seeing if that fixes it?

If it does not fix it, could you share you calibration files so that I can try to reproduce the error?

Thank you!
Pierre

[menor22]

Hi Pierre,

So I tried to upgrade calligator but I got the same error message.

Please find attached my config file (in .txt since I can't join .toml files) and videos for calibration.

Thanks for your help!

Yannick

Calib_vidz.zip

config.txt

[lambdaloop]

Hi @menor22 ,

I tried reproducing your issue, but unfortunately, I cannot. I seem to get slightly different detections and then it runs fine... I tried running it with both OpenCV 3.4.7 and OpenCV 3.4.5 .

For reference, here is what I get as output:
https://drive.google.com/open?id=1aNiSSCanT6Aujlm6-Ct9UXO_-xv4iadq

I suspect there must be some kind of invalid detection that throws it off when it tries to estimate the pose. Could you share your detections.pickle file? Perhaps I could find which one it is to patch calligator. Feel free to zip if it needed to share it.

Thank you!
Pierre

[menor22]

Ok, so it was probably a problem with my video files (too big? >1 Gb). I tried with the one I sent you (compressed) and now it's working fine.

I need a reference frame but I don't really understand the matrix in the config file:

[triangulation]
axes = [
    ["x", "L1A", "L3A"],
    ["z", "L1B", "L1A"]
]
reference_point = "L1A"

What should I change for my camera settings?

Best,

Y

[lambdaloop]

Hi @menor22,

Glad that it is working well! If you still have the detections.pickle file for the big video files, I would still appreciate seeing it, so that I can fix the bug you encountered.

For the reference frame, the documentation for it is here:
https://github.com/lambdaloop/anipose/blob/master/docs/start_3d.md#configuring-the-standardized-3d-pose

Depending on your setup, you might label additional stationary points to use for a reference frame.

--Pierre

[menor22]

Hi Pierre,

Please find attached the detections.pickle file.

So if I understand well I have to label (from deeplabcut) fixed points to use them for reference frame?
I guess that using bodyparts for reference frame is not a good idea (e.g. ['x','shoulderR','wristR'], ...)?

Best,

Y

detections.zip

[Edit]

I tried to calibrate 3 cameras (3 .avi files with exact same number of frames) and I got the same error message as I got previously. I tried different compression, etc, but I still have this error message.

[lambdaloop]

Hi @menor22 ,

Thank you for providing the detections! I was able to reproduce your bug. I'm not sure why, but for some reason on some frames there were some checkerboards detected with the invalid number of points. I added a check for that in calligator and it seems to work now.

Could you update calligator and try again?

For reference frames, anipose sets a fixed reference frame per video, using the median values for the point locations. If your points do not move much or move around some location periodically, then it should be fine to use them for reference. If they move around too much, then with the current algorithm the reference frame may not mean much.

Thank you!
Pierre

[menor22]

Hi Pierre,

So I updated calligator and it works now, thanks a lot.

For the reference point it's not possible to define a fixed value by providing x and y coordinates (and same for the axis) instead of add a new label in deeplabcut?

Best,

Yannick

[lambdaloop]

Hi Yannick,

Great! I'm glad this fixed it. I will close this issue then, but feel free to comment or make a new issue if you have any other issues.

Unfortunately, anipose doesn't currently support a reference frame with static x and y coordinates per image. Perhaps you could file an issue for this, I can get around to it eventually.

Best,
Pierre

[menor22]

Hi Pierre, Sure, I will work on it and let you know when it’s done. Thanks again for your help. Best, Y