Latest Samples
Corrected-test sample montage from the newest training pass: raw | corrected GT | prediction | TP/FP/FN error. Refreshed after each pass.
Open Samples →● permanent — refreshed after each training pass
Vision-LLM Experiment
Vision LLM vs our pipeline on 10 training images: segmentation Dice/segR8 plus FP/FN adjudication (Gemini 3.7 Flash, stealth/ox-alpha). Static, read-only results.
Open Experiment Results →● static — regenerated per eval run
VLM Benchmark
Zero-shot coronary segmentation: 10 current vision LLMs vs the frozen GT on 10 random ARCADE images. Skeleton recall/precision + Dice, per-model overlays. Pilot (n=10).
Open Benchmark →● static — regenerated per benchmark run
Extras Annotator
Judge model-extracted segments one at a time: TRUE (real vessel missed by ARCADE) or FALSE (artifact). Builds the confirmed-extras DB.
Open Extras Annotator →checking…