0
$\begingroup$

Current - frustrating - progress

I'm continuing to work my way through the original Match, Track, Blend(MTB) DVD, but I've smacked into a baffling problem in Chapter 16: Creating a Clean Plate.

As you can see in the image above, I've tried to replicate as closely as I can the markers that Sebastian Koenig chooses for his tutorial (I've also tried to copy his camera settings).

However - my camera solution is often crazy, varying between 8 and 240. Not changing any variables and just clicking on Solve Camera Motion will usually return a new, often very high value.

Shortly after 6.40 in the video, Sebastian sets the solver to refine K1 & K2 and while I've tried the same, it only seems to restrict the average error within 16 units or so.

Obviously MTB was made for an earlier version of Blender - but could there be settings from an older version which are more effective than the current Tracker defaults? And if so - what are they!? I've certainly experimented with affine, etc.

Help and advice, very much appreciated. The DVD has been mostly fantastic, and I've learnt tons... just a shame to hit a brick wall on this useful lesson!

--Rev

PS: Bonus question! Obviously I haven't finished the Clean Plate tutorial, but I took a look at this vid - https://www.youtube.com/watch?v=s_MXkTP73CI Is there a chance the plate cleaning / clone painting tech has made it into Blender yet?

$\endgroup$
  • 1
    $\begingroup$ In general we don't really like bonus questions, unless they are close related or dependent on each other. Please write only one question per post to keep things organized... The tools that Sebastian Koenig is talking about in that video are scripts he's developed for his own workflow, that might or might not make it to blender... The functionality is already in blender, but is not a simple button or a tab. $\endgroup$ – user1853 Jul 12 '15 at 21:07
  • $\begingroup$ Apologies, noted, will avoid in future. --Rev $\endgroup$ – Reverend Speed Jul 13 '15 at 10:44
2
$\begingroup$

Most errors will be fixed by inproving the trackers' accuracy. There are a few tools you can use:

On the tracking window's dopesheet sort your trackers by average error. This will show you which trackers are not tracking correctly.

enter image description here

On the graph window look for places where the graphs are clearly different than the rest.

enter image description here

To fix those issues Retrack, reposition, resize, affine, use a different kind of motion model algorithm or tracking preset or do whatever you can to improve the accuracy of those trackers to to bring down the error average..

| improve this answer | |
$\endgroup$
  • $\begingroup$ read also: blender.stackexchange.com/questions/8934/… $\endgroup$ – user1853 Jul 12 '15 at 20:45
  • $\begingroup$ I'm aware of these techniques, but strongly appreciate the input. However, as stated, I'm following SK's chosen markers and settings as closely as possible - and (assuming tut is not edited so efforts to correct grievous errors are excised) he seems to have little trouble getting a decent solve on a series of straightforward tracks (w/excellent source footage). I'm not really looking for a way to massage my current data, more if anybody can spot a flaw or divergence in my technique from that shown in Sebastian's video (or if there's a setting I can change to recreate his version of Blender! $\endgroup$ – Reverend Speed Jul 13 '15 at 10:50
2
$\begingroup$

Ultimately the solution was provided by sebastian_k himself on the blenderartists.org forum, quotation follows:

Yeah, that was a bit of a weird problem. Lots of good markers, seemingly enough parallax (a moving camera) etc. So why did it fail? As often, the main tool to get a good solution is the placement of the keyframes. Unfortunately the automatic keyframe detection doesn't always do a good job on that. But here's what I did: Actually it did help me to open up the scene without the footage, because i could examine the movement of the markers. And what I found by scrubbing through the shot was that I could see the most parallax around frame 75 and 110. So setting keyframe A to 75 and keyframe B to 110 already gives you a solve error of 1.4, which is a nice starting point to bring it below one using Refinement for focal length and K1 and K2. So, if you have troubles solving a scene and are confident your markers are good, always try different keyframe settings. Set keyframes when you see the most parallax in your shots.

Following this advice with the tutorial footage indeed ultimately resulted in a 0.6544 solve - and the tutorial could continue!

It's perhaps worth noting that this still doesn't explain why (almost! very! nearly!) identical actions across two different versions of Blender produced such strongly divergent results. Something to be aware of, maybe...

(Mumble mutter bonus question answer: https://github.com/sebastian-k/scripts/blob/master/cleanplate_creator.py never again)

| improve this answer | |
$\endgroup$

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.