I feel both blessed and cursed to have discovered this repo, as it does something I've been desperately wishing for for several projects I've been working on the past year or so, but it requires so much manual intervention that it's a pandora's box of sorts. Knowing AutoOverlay exists, these projects can become a reality, but would involve many hours of manual work.
So before I sink my teeth into the visual editor, is there any way to make AutoOverlay behave less erratic within a scene? For instance, I'm working on aligning the newly remastered version of The Matrix with the original theatrical version which has much warmer colours. Within the first scene (cops chasing Trinity), AutoOverlay chooses wildly varying alignment coordinates, which are quite obviously wrong, while still claiming the frames are "successfully aligned". If it had just stuck with the alignment values of the preceding and/or following frames, it would have been fine. Tweaking the max difference values, branches, correction, or subpixel precision all don't help, but only make the process arrive at the same result much, much more slowly.
I feel both blessed and cursed to have discovered this repo, as it does something I've been desperately wishing for for several projects I've been working on the past year or so, but it requires so much manual intervention that it's a pandora's box of sorts. Knowing AutoOverlay exists, these projects can become a reality, but would involve many hours of manual work.
So before I sink my teeth into the visual editor, is there any way to make AutoOverlay behave less erratic within a scene? For instance, I'm working on aligning the newly remastered version of The Matrix with the original theatrical version which has much warmer colours. Within the first scene (cops chasing Trinity), AutoOverlay chooses wildly varying alignment coordinates, which are quite obviously wrong, while still claiming the frames are "successfully aligned". If it had just stuck with the alignment values of the preceding and/or following frames, it would have been fine. Tweaking the max difference values, branches, correction, or subpixel precision all don't help, but only make the process arrive at the same result much, much more slowly.