Six Models Fail to Reproduce Thompson's Animal Morphogenesis Plates as Benchmark Draws Criticism
An evaluation asking image models to recreate D'Arcy Thompson's classic morphogenesis plates found all six models failed, scoring no better than a featureless color-block baseline. Critics argue the benchmark's scoring overweights edge alignment over perceptual similarity and contains multiple questionable design choices.
2026-11-06 ~ 2026-11-06 · 3 related posts
- Six AI models fail to replicate Thompson's morphology plates, losing to a featureless blob baseline — giffmana · 2026-11-06
- Benchmark critique: shadow-puppet score favors edge alignment, zero of 6 models solve a plate — cloneofsimo · 2026-11-06
- Chollet and critics clash over image-gen eval favoring edge alignment over perceptual similarity — giffmana · 2026-11-06