Skip to content

Dataset / solver notebooks

Demo and analysis notebooks for the data-generation side. Each page links to the source notebook, which you can open on the Hugging Face Hub or download and run locally.

Notebook rendering and licenses

  • Getting started

    The shortest path from the dataset to a training example: load one released scene, decode its tensor, draw a reference route and turn it into a segmentation target.

    Get started with PASSAGE

  • Obstacles

    Obstacle masks from least to most covered, drawn coverage against its target, and the detour that obstacles force on the optimal route over a whole validation split.

    Explore obstacle coverage

  • The three objectives

    What the elevation, slope_uphill and energy objectives reward on one scene, their elevation profiles, and the extra cost of a route optimised for the wrong objective.

    Compare the three objectives

  • Exact re-solve

    Solves stored labels again with the exact float64 A* reference, compares costs and routes, shows why the search budget matters, and repeats the check on a sample of validation scenes.

    Check the reference routes

  • Timing a planner

    Times the canonical A* planner of passage-eval with the practices that make a timing meaningful: warm-up, repetitions, percentiles and a stated machine.

    Time a planner