Controlled benchmark of Vision Transformers for cervical-vertebra landmark localisation in videofluoroscopy. 169 archived runs collapsed into a 9-model x 6-configuration grid, 95% CI on every number, each re-derivable from a single run hash. 243 tests; no GPU or dataset needed to verify.
benchmark deep-learning reproducible-research pytorch medical-imaging landmark-detection pose-estimation vision-transformer vitpose videofluoroscopy
-
Updated
Aug 5, 2026 - Python