I had the same idea for 3D-to-2D conversion, but I was unable to solve the image distortion problem after splitting the video into two sides. As a result, AI models that work on standard videos/images are ineffective. Many years have passed, and I was wondering what your current progress is.