SeamCam: Quantifying Seamless Camouflage via Multi-Cue Visual Detectability
arXiv, 2026
Animals are called well camouflaged when they blend seamlessly into their surroundings, yet no standardised quantitative measure of that seamlessness exists. SeamCam frames camouflage evaluation as visual localisation: a well-camouflaged animal stays hard to detect even when its category is known. In a two-alternative forced-choice study with 94 participants and 2,390 comparisons it agrees with human camouflage judgments 78.82% of the time, roughly 25 points above the state of the art, and doubles as a preference signal for fine-tuning camouflage generation.