0
$\begingroup$

I am using this as a last resort, since I have been reading forums and trying over and over again, and have watched just about every YouTube tutorial. Blender is giving me such random results on my tracking that it appears to just be absolutely broken.

I printed and used some good tracking sheets. I mean, from where I was before with tape, the tracks, visually anyhow, look very clean. When I scrub through, it pretty much stays within 1 or 2 visible pixels. enter image description here

So at first when I solved, i was getting about 1.7 average error, which i have, with much frustration, been trying to improve upon. But I can stay at the same settings that I got the 1.7, and it swings wildly up to 200 average error, to erroring out saying it cannot reconstruct. Literally without changing a thing, or, changing and going back to the previous setting. Won't ever go back to the 1.7.

I have stabilization off, set at 1920x1080 60fps, video frames rendered to an image sequence as EXR, have tried a lot of different ways here, and I still just get absolute randomness. Video made is from a Pixel 4XL, and I hear "reviewers" talk about how awful the videos from the phone are, but man it seems pretty solid visually to me, so I will try using my SLR full frame camera and see if all these issues just go away, or if i still get complete randomness from Blender. Anyone else seen these wild swings in results and know of general ideas of why Blender reacts this way?


Thanks for the input. I do not find the answer helpful, however I do appreciate you took the time. A solid track visually means it is following the same point, even down to a pixel or two. I have used tracking a while not, and in Blender 2.7x I was pretty good at getting the results I needed.

I recorded the video again, only with an SLR full frame sensor and 17MM lens, not on a tripod. I literally didn't have to keep adjusting the tracking points as it was able to follow the tracks with ease, even though VISUALLY it looked the same as the other video at the zoomed level. There is clearly much more detail in the videos than Blender leads you to believe in the rendered preview of the video.

However, while i do not expect an actual answer to my question on Blender being broken, my follow on video that I had success with still suffered the same random wild throws in camera solve. I was getting 80.72 average on everything (16 points), even though it tracked phenomenally, no matter what settings I changed. Then, all of a sudden, when I changed the settings back to what I had started with (auto keyframe, refine k1 k2), which is what I expected would work in the first place, the camera solve jumped from the constant 80 to a solid .32 after just pressing solve camera over and over out of frustration. Out of the blue, really did nothing but kept trying to push solve camera, and wham. Nailed it at .32. So, again, i am not sure how to report bugs, but I will look into it.

$\endgroup$
1
  • $\begingroup$ Blender is not broken and nothing changed in that particular area since 2.7x. As mentioned in the answer, while essentially having nothing to work with, hard to give good advice based on your question. Anyway, if you want to report a bug just go to the Main menu > Help > Report a Bug... $\endgroup$
    – brockmann
    May 11 '20 at 18:58
2
$\begingroup$

Blender's trackers and 3D reconstruction are quite robust and advanced, but it is hard to pin down what is happening without access to the video you are trying to track. "Solid visually to me" can mean anything or nothing at all.

Solid reconstruction is only possible with accurate camera tracking.

Common mistakes are:

-Trying to track blurry or distorted video, or video where the tracked elements move to far from frame to frame, for that you need to enlarge the search box.

-Expecting 3d information out of shots where the camera is stationary (where the video was taken from a fixed place and the camera just pans and tilt without any displacement). For those shots you can use "Tripod" Solve, but know that there is no way to get 3d information out of such shots. It is considered 2D tracking as you will only get the camera movement but no real 3d reconstruction.

There is a very detailed post on this site that you should read to troubleshoot your problems.

How can I get better results when doing camera motion tracking?..

$\endgroup$
2
  • $\begingroup$ This is a pretty good answer, but linking to outside sites is bad form. Can you inline the parts you find useful to this answer? $\endgroup$
    – Kirbinator
    May 8 '20 at 17:41
  • 2
    $\begingroup$ @Kirbinator The useful parts are already in the post. No "outside links" to other sites on this answer... please read carefully. I won't go into repeating information that is already accessible and carefully written. As a matter of fact this whole post is redundant with the linked question and answer. Maybe someone can mark it as duplicate. $\endgroup$
    – no-can-do
    May 8 '20 at 17:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.