- Align metadata with AI Portal schemas (7a1357df44cb15ef876fffc9af456d5d89efe738)
- Use the current dataset measurement (e0976c16c1681772a8ce3771cd399b096b7e4ede)
- Generate model weights locally (36d3072bf033778f5e19d4444805ab5fd348f942)
- Add temporary model placeholder (fb9954e72cc3df595907add41ee6785cce9fa9d6)
- Remove model placeholder (8033b735257221adca5752d92ca888ff3ef94a05)
- Limit PR to catalogue metadata (b00f44ec6782af99774d7df7acf222e03a2c1394)
- Address model metadata review (74d913a37b61e1814b3514b4c5b3a4bcd991b483)
- Pin accuracy to published dataset revision (41c77efb0559267f940e7ef228d8eacf1b0bdc02)
- Use catalogue runtime name (e03e24fef6abb90b18ee5229d9ca7b67163d8c7e)
Replace the earlier prototype package with the public runtime layout, converter and Python API; add the patched Ultralytics benchmark workflow, licensed validation dataset integration, fresh Marvik metadata, and verified reference results. The PyTorch checkpoint is no longer redistributed.
