densemaps

densemaps: abstract, memory-scalable correspondence maps for 3D geometry.

Two interchangeable backends are provided as subpackages:

• densemaps.numpy – NumPy/SciPy, CPU only, no PyTorch required.

• densemaps.torch – PyTorch (CUDA/keops) backend; adds the memory-scalable KernelDistMap. Importing it requires torch (install the [torch] extra).

The top-level package intentionally does not import the torch backend, so that import densemaps works in a torch-free environment.

Modules

numpy	NumPy/SciPy backend for densemaps (CPU only, no PyTorch required).
torch	PyTorch backend for densemaps (CUDA / keops capable).