Seems a bug in estimate_pose_points

[Eddylib]

multical/multical/board/common.py

Line 42 in 76b8987

valid, rvec, tvec, error = cv2.solvePnPGeneric(objPoints, undistorted, camera.intrinsic, camera.dist)

In this line undistorted points are input to cv2.solvePnPGeneric, the distortion should be zeros: np.zeros(camera.dist.shape).

In my fisheye dataset, the current version, cv2.solvePnPGeneric leads to average error of 18.69. When np.zeros(camera.dist.shape) are input to cv2.solvePnPGeneric , average error becomes to 3.44

[oliver-batchelor]

Yes I think you're absolutely right, this looks like it is undistorting twice. Possibly I never noticed because I didn't test with high distortion cameras and it still find an OK initial solution...

…

On Thu, Dec 5, 2024 at 3:53 PM BaoyuLi ***@***.***> wrote: https://github.com/oliver-batchelor/multical/blob/76b8987d4851af44c83cee2a03abacd4fde2d7cf/multical/board/common.py#L42 In this line undistorted points are input to cv2.solvePnPGeneric, the distortion should be zeros: np.zeros(camera.dist.shape). In my fisheye dataset, the current version, cv2.solvePnPGeneric leads to average error of 18.69. When np.zeros(camera.dist.shape) are input to cv2.solvePnPGeneric , average error becomes to 3.44 — Reply to this email directly, view it on GitHub <#73>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAITRZPTOGQLFPUOOJAX57T2D653NAVCNFSM6AAAAABTBPSHA6VHI2DSMVQWIX3LMV43ASLTON2WKOZSG4YTSMJYG42TAMA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[Eddylib]

And there might be other issues in code for fisheye data. In my fisheye dataset, the code outputs very large reproject error during optimize, and will stop immediately.

[oliver-batchelor]

If you find out why, I'll be happy to accept a pull request. I don't currently use this tool myself, and the fisheye code was supplied by a 3rd party so I haven't used it previously!

…

On Thu, Dec 5, 2024 at 8:46 PM BaoyuLi ***@***.***> wrote: And there might be other issues in code for fisheye data. In my fisheye dataset, the code outputs very large reproject or during optimize, and will stop immediately. — Reply to this email directly, view it on GitHub <#73 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAITRZO6QGG4NYIS4V2MQRL2EAAFBAVCNFSM6AAAAABTBPSHA6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKMJZGQ4DKNZXGA> . You are receiving this because you commented.Message ID: ***@***.***>

[Eddylib]

Okay, I'll take some time to figure it out. Thank you for your reply and your excellent work.