Major consumer-wide video stabilization programs zoom/crop to readjust for the camera movement to get to a stable footage. But that results in loss of content and data.
We work on each frame individually. We extract all the frames, and recompensate for the lost data while stabilization using outpainting, diffusion models and extrapolation.
After stablizing a video with moviepy, vidgear there are black edges that represent lost data. Our task was to fill that sections on each frame. We solved it by using diffusion.
INPUT VIDEO:
STABILISED VIDEO WITH PRIMITIVE STABILISATION:
STABILISED VIDEO WITH FILLED EDGES:
FPS interpolation to increase frame rate to a defined rate.