Paper-inspired benchmark surface

A live benchmark surface for cross-domain medical TTA

Loading benchmark summary...

Browse methods Open modality drilldown Jump to local code

MedSeg-TTA paradigm overview figure — The repository taxonomy figure stays visible so the leaderboard never loses the benchmark's four-paradigm frame.

MVP surface

What this first leaderboard already does well

The page exposes the benchmark at three levels: paradigm trends, overall method ranking, and modality-specific drilldown. Local implementations also keep direct GitHub jumps so the board stays connected to runnable code.

Rank colors Best Second Third Below target-domain baseline

Use the view switcher for the main leaderboard surfaces, then scroll further for direct repository jumps into the local method folders.

GitHub jump map

Local implementations organized by paradigm

Every method with local code links back to its canonical repository folder, with dimension-specific jumps for `two_d/` and `three_d/` where available.

Why this shape

Academic tables, turned into a workable first release

The interface keeps the visual semantics from the benchmark figures: paradigm symbols, top-three cell highlights, baseline-aware gray rows, and modality-specific drilldown. This MVP focuses on the benchmark's primary metrics, Dice and HD95, so the board stays readable while still feeling rigorous.

MedSeg-TTA benchmark framework figure — The framework figure remains visible here as context for how leaderboard entries map back to the benchmark setup.