wanghlsara
commented
1 month ago In order to increase the number of iterations of optimization, I used following code in test_calibration.py
quality_controller = QualityController(capture_volume, charuco)
for _ in range(5):
logger.info("Filtering out worst fitting point estimates")
quality_controller.filter_point_estimates(0.25)
quality_controller.capture_volume.optimize()and got python exception:
File "C:\WORK\caliscope-main\tests\test_calibration.py", line 140, in test_post_monocalibration
quality_controller.capture_volume.optimize()
File "C:\WORK\caliscope-main\caliscope\calibration\capture_volume\capture_volume.py", line 93, in optimize
self.least_sq_result = least_squares(
File "C:\Github\00.caliscope\.venv\lib\site-packages\scipy\optimize\_lsq\least_squares.py", line 832, in least_squares
f0 = fun_wrapped(x0)
File "C:\Github\00.caliscope\.venv\lib\site-packages\scipy\optimize\_lsq\least_squares.py", line 830, in fun_wrapped
return np.atleast_1d(fun(x, *args, **kwargs))
File "C:\WORK\caliscope-main\caliscope\calibration\capture_volume\capture_volume.py", line 162, in xy_reprojection_error
points_3d = current_param_estimates[capture_volume.point_estimates.n_cameras * CAMERA_PARAM_COUNT:].reshape((capture_volume.point_estimates.n_obj_points, 3)
)
ValueError: cannot reshape array of size 93213 into shape (31065,3)I've found that after each iteration of optimization, the capture_volume.point_estimates.n_cameras may become smaller. If I write it as a magic number (equal to the number of videos) in the code, Exception won't happen again.
But it doesn't seem to make sense to modify like this..... So,Looking forward to your suggestions.
mprib
I mentioned in #659 that I performed extrinsic calibration in a narrow and long rectangular area. I copied some lines related to rmse from calibration.log. Please forgive me that these lines may not be continuous because I ignored the information unrelated to rmse. What confuses me is that the rmse of each port pair looks good, but the overall error is large. What could be the reason? Is it because there are few cameras with overlapping views in the rectangular area? Or is the ratio of FILTERED_FRACTION=0.025 too small? Or is the number of bundle adjustment stages too small? What aspects should I investigate next? I hope to get your advice.