torchmatch

Resources

Material that cuts across both solver families — interactive tutorials, application history, and the benchmark sweep.

Material that applies across both solver families, rather than to just one:

  • Tutorials — interactive Jupyter notebooks covering both families from first principles.
  • Assignment applications — where the linear assignment problem shows up, from 1950s operations research to DETR and multi-object tracking.
  • Transport applications — where optimal transport shows up, from Monge's original problem to Wasserstein GANs and single-cell genomics.
  • Benchmarks — per-op sweep across problem sizes, dtypes, and devices, covering assignment and transport ops alike.

See also the API reference and the unified bibliography for every citation used in the documentation.

Point clouds

Computing OT directly on raw point sets with samples.loss, training a neural network shape generator, handling outliers with unbalanced OT, and computing Wasserstein barycenters.

Tutorials

Interactive Jupyter notebooks covering both the assignment and transport families, from first principles through applied examples.