Right, I put together a comp to compare all the different methods. And the results are.. a bit confusing.
Methods compared:1 & 2: Planar Tracker FG only vs Planar Tracker FG over BG = identical.
3: Corner Positioner -> Planar Transform = subtly but noticeably different to 1 & 2, with the Planar Transform result looking blurrier (possibly noticeable to human eye)
4: PTra Steady -> Corner Positioner -> Planar Transform -> PTra Unsteady = almost identical to 3, but with very subtle differences, unnoticeable to the human eye. As this is very similar to 3, it would therefore be more noticeably different vs 1 & 2.
5: ImagePlane3D into PlanarTransform = I couldn't get this to match due to AR differences, discussed below. I may well be doing something wrong here, if it's working right for you.
6. ImagePlane3D into Planar Tracker in FG over BG mode = ditto
Comparison composition:I've made a test comp that explores all of these options, using some stock footage and a Test Card image in three different aspect ratios (16:9, 1:1, 3.5:1).
It also includes Merge nodes in Difference mode so you can quickly compare the difference in output between the various methods. The reference frame is 57, which is noted in the comp.
It includes the source footage and is 40MB, which is too big to upload as an attachment here, so you can download it from:
-
Google Drive link.
Fusion Studio vs Resolve StudioI made it in Fusion Studio, and if you unzip it and then open the .comp in Fusion Studio, it will all work immediately as it will load the media from the same directory as it finds the comp.
If you've not got Fusion Studio set up, you can instead make a Fusion Composition in Resolve, go into the Fusion page, then File -> Import -> Fusion Composition. However this will not automatically find the media. You then need to go to each Loader and load the appropriate PNG file. Each PNG is in a group named appropriately, eg TestCard169OverBG so you can see which file that group is meant to be loading:
- TestCard169OverBG = TestCard.png
- TestCardWideOverBG = TestCard3to1Wide.png (actually 3.5:1)
- TestCardSquareOverBG = TestCardSquare.png
The video footage can't be loaded by MediaIn in Resolve, so you'd need to drag it from the filesystem into Resolve, creating a new MediaIn, which you can then connect to the two PipeRouters that come off the "PexelsBuilding" node.
But it's really a lot easier to do in Fusion Studio as it'll just work immediately.
Discussing the comparisons:Short answer is I don't have much experience in this area and I don't really know why there are these differences or what the recommended method is.
I feel like the difference between the Planar Tracker vs Planar Transform is something to do with filtering, but I don't know the specifics. There are no options I can find in the Planar Transform to control the scaling filtering, which is something one would see in a Merge, Transform or Resize node.
It's also possible that the difference is caused by the Corner Positioner rather than the Planar Transform.
I should add that I'm testing this on a test card image which is explicitly designed to clearly show differences, by virtue of containing large squares of colour and narrow thin lines. Whether the differences would be so apparent in a real life image, I'm not sure.
Based on these results, it does seem like Planar Tracker Corner Pin (either FG only or FG over BG) is the best option, at least compared to a Planar Transform.
Going into 3D:I've never heard of anyone using the ImagePlane3D method in a 2D tracker - it's a common method for 3D camera tracks though.
I can't see any fundamental problems with it, besides a reduction in performance due to going into 3D with a Renderer3D. I think that if it gives the result you want, it should be fine to use it.
However in my tests thus far it's always giving me an altered aspect ratio. At first it was completely wrong, but then I realised that for the ImagePlane3D method, it's necessary to use the original image -at its original aspect ratio - and not the image merged over a frame size background.
However even that isn't exactly right. I still see AR adjustments. If I change the Renderer3d to have the correct aspect ratio - eg 2160x2160 instead of 3840x2160 - it looks much closer, but then it doesn't stick correctly across the full frame range.
So I'm a bit bemused about what you're doing differently there. But quite possibly I'm misunderstanding something or setting it up wrong.
I'll leave this here and upload my test comp and wait to see your example of how you're doing it.
3D Camera Tracking:The use of ImagePlanes is commonly used in full Camera Tracking. You may already be aware of this, but in case not, here's a good tutorial on how this same sort of process could be completed in full 3D: