WildRoadBench - Road Crack: leaderboard

The road crack class of WildRoadBench (2026), an aerial road-damage grounding benchmark: models box every instance in wild drone images, 561 ground-truth boxes in all. The score is F1 at IoU 0.5 in percent; higher is better.

Metric: F1 (%) of predicted boxes at IoU 0.5 for the road crack scene class (561 ground-truth boxes), VLM Track, zero-shot box grounding on the full 1,061-image aerial set (1,394 ground-truth boxes); higher is better. Source: arxiv.org. Saturation forecast: Around October 2027. 25 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3 Pro33.3#146
2Qwen 3.6 Plus25.4#245
3Gemini 3.1 Flash Lite16.2#377
4Qwen 2.5 VL 7B Instruct15.3#1919
5Qwen2.5-VL-32B-Instruct-AWQ11.7
6Claude Sonnet 4.611.3#179
7GPT-5.4 (2026-03-05)10.8#95
8Qwen3.5-VL-27B-FP87.4
9Qwen3.5-VL-35B-A3B-FP87.3
10Qwen3-VL-32B-Instruct-AWQ6.5
11Qwen 3 VL 8B Instruct6.2#1043
12Qwen 3.5 VL 9B5.6
13Qwen3-VL-32B-Thinking-AWQ5.1
14Qwen 3.5 VL 4B4.3
15Kimi K2.54#272

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=wildroadbench-road-crack · How It Works · Data refreshed daily, snapshot 2026-11-22.