The problem in Issue #41 was caused by a mismatch between the number of frames in the video and the length of the frames list input into the make_reproj_imgs() function. In my tests, I noticed that earlier sample frames seemed to match, while later sampled frames did not match. This is because each of the frames that lead to no detections would be subtracted from the list of total frames. So they become progressively more mismatched. The good news however is that the calibration itself was not compromised by this bug just the plotting. So calibration files should be good to go, and histograms remain an accurate representation of the error.
I have written a new version of the make_reproj_imgs() function which grabs the frame from the video using imageio and ffmpeg rather than from a list of images. The eric/calibration-imgs branch eliminates the problem of misaligned detections, reprojections and images described in Issue #41.
The problem in Issue #41 was caused by a mismatch between the number of frames in the video and the length of the frames list input into the make_reproj_imgs() function. In my tests, I noticed that earlier sample frames seemed to match, while later sampled frames did not match. This is because each of the frames that lead to no detections would be subtracted from the list of total frames. So they become progressively more mismatched. The good news however is that the calibration itself was not compromised by this bug just the plotting. So calibration files should be good to go, and histograms remain an accurate representation of the error.
I have written a new version of the make_reproj_imgs() function which grabs the frame from the video using imageio and ffmpeg rather than from a list of images. The eric/calibration-imgs branch eliminates the problem of misaligned detections, reprojections and images described in Issue #41.