waxmorph.torch.losses

PyTorch losses for row-aligned or unordered point clouds.

squared_loss requires row correspondence. Chamfer and GeomLoss samples are permutation-invariant; JAX offers OTT Sinkhorn, while PyTorch offers GeomLoss families.

Functions

chamfer_distance(X_pred, X_target)	Sum of directional mean nearest-neighbor distances.
make_samples_loss([params])	Return a lazily imported GeomLoss loss for unordered clouds.
squared_loss(X_pred, X_target)	Squared Frobenius loss for row-aligned clouds.