2D or 3D scene flow estimation?

[TedLentsch]

I checked the demo notebook and the MLP is used to predict a 2D flow, i.e. the Neural_Prior class is initialized with dim_x=2. However, the function optimize_neural_prior in the script optimization.py creates a class instance without providing the dim_x argument, i.e. the default of dim_x=3 is used, so 3D flow is predicted. Therefore, which setting should be used according to the authors?

[Lilac-Lee]

Hi, @TedLentsch, scene flow describes the 3D motion field, therefore, our project targets 3D point cloud-based scene flow estimation. The demo notebook is just a 2D example for better visualization and understanding. Cheers.