DefectBench (Building Facades) - Generative Segmentation: leaderboard
Metric: Mean IoU (0-1, times 100) of crack masks produced by image-editing models told to repaint a facade image as a binary crack mask, after adaptive binarization and region-of-interest gating; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Nano Banana Pro (Gemini 3 Pro Image) | 74.18 | |
| 2 | FLUX 2 Pro | 69.26 | |
| 3 | Seedream 4.5 | 69.2 | |
| 4 | DefectBench Qwen-Image-Edit-Plus (checkpoint unspecified) | 68.52 | |
| 5 | Wan2.6-image | 68.48 |
Interactive version: theaggregate.ai/benchmark?slug=defectbench-building-facades-generative-segmentation · How It Works · Data refreshed daily, snapshot 2026-10-11.