Page MenuHome

improve usability of 2D stabilizator
Closed, ResolvedPublic


While Blenders video stabilizer, based on the great image feature tracking support, works well, is easy to use and fulfils all basic needs, its usability on more complex shots and extended real world footage is still somewhat lacking and might be improved...

  • adding more than two tracks does not improve the result in most cases and does not help to get the sometimes unpredictable bumpiness under control, even when adding carefully chosen tracking points, which in theory should improve the result.
  • the canvas does not adapt and expand on demand, which makes it near impossible to deal with sidewise tracking or panning shots
  • tracks with later onset and gaps cause hard to understand problems
  • rotation stabilisation works quite well in a “clean demo” setup, but is unusable when it’s needed most.
    • only a single track can be used to control rotation
    • this track doesn’t pick up the angular movements in a correct way
    • when the tracking marker moves towards the top or the bottom, above or below the other tracking points, unpredictable jumps can happen.
    • when some tracks start at differing positions, the rotation compensation flounders completely, flipping the frame upside down
  • while it is possible to animate (some) properties of 2D stabilization, it is not possible to open F-curve or dopesheet editors on such animation data beyond the scope of the main 3D scene.
  • sometimes it would be desirable to control the actual effective rotation centre (pivot point), which is not necessarily fixed to the image centre. There should be the option to anchor this either to the weight centre of the tracking points (default), the image cadre or the canvas.
  • controlling the effective camera movement is possible, yet laborious and often frustrating. In many cases, a simple lowpass damping would resolve most of the problems and leave only the fine adjustments to the user.
  • stabilzation creates "dancing black borders", forcing us to zoom in. Too much zoom decreases image quality and kills image composition. Often, it would be possible to fill those black border gaps with footage from adjacent frames, yet doing so manually is a tricky and laborious task

Event Timeline

Aaron Carlisle (Blendify) lowered the priority of this task from Needs Triage by Developer to Normal.Aug 8 2016, 10:22 PM

after some testing, tweaking and further investigations, the version (rBbaaa2 + rB9c3b9) currently in master changed the pivot point to follow the weight centre of the location tracks, instead of attaching it to the image centre. The pivot point is the reference point to detect rotation and scale against, and thus this change improves the behaviour especially in very simple situations. For example, if you add only a single (location) track, this location also becomes the pivot point. If you then add just one other marker as rotation track, the stabiliser will keep the relation between those two points fixed, and scale/rotate the image accordingly to do so.

Aaron Carlisle (Blendify) closed this task as Resolved.Dec 25 2016, 4:43 AM

AFAIK everything has now been committed and this can be closed.

Hello Aaron,

actually I created this task as an umbrella for that topic; I have still some things to improve and I intend to contribute more to that topic. Note especially the point with the "dancing black borders" is not solved. Also the handling of the pivot point can be bettered (and I know how). And, some people proposed different approaches to the same problem, like using a lowpass filter on the track data

Only the immediate Diff D583 is closed and released, which resolved the most pressing issues.

Indeed I am unsure how such an ongoing effort is managed in Blender. Someone recommended me to create such an umbrella task. But I do not mind if you close it, just lets's make it clear in the text then, that it is only closed for technical reasons, not really resolved. An I'll just create new diffs when I have something to contribute again.