This is a research repository aimed at studying the possibility of doing 3D reconstructions using computer vision along with AV1's video codec compression data (such as motion vectors). An overview can be found here. A proper paper is under way.
For the source code check source code, for the TODO list check todo, for the online sources sources.
In order to export compression data from libdav1d and FFmpeg I'm building modified versions of them at libdav1d FFmpeg as they don't export them.
A list of AI-generated code is available here, along with the reasoning for using it. All code, after generated was reviewed by me with care.