Kvasir-SEG — Polyp (endoscopy)

Deployed pipeline: op_sam dense-correspondence localizer -> SAM -> genome -> reprompt keep-best — test Dice 0.8646 (this cache: mean 0.8646 / median 0.9484, n=100).

▮ green = ground truth    ▮ red = deployed prediction. Per-image Dice is burned on each tile; 6 fixed test images, chosen deterministically (spread across the dice-sorted test set).

cju2zy1e49pqk0878t6ncqn12 · Dice 0.005

cju8b3ka8r64u0801fh18hk7l · Dice 0.767

cju5wkonqlrl409877y8zvnub · Dice 0.917

cju5yimthmlv80850zhoc90c2 · Dice 0.955

cju85l4yjlops0801fvmnwptf · Dice 0.973

cju2u2b9o4zvp08788qb9nqxj · Dice 0.989