AgroVG - Multi-Target Set-F1: leaderboard

AgroVG (2026) agricultural visual grounding on queries naming several targets in dense field imagery: the predicted box set is matched to the true set at IoU 0.5. Query-averaged Set-F1 in percent; higher is better.

Metric: Macro Set-F1 at IoU 0.5 on multi-target queries (%, times 100), zero-shot agricultural visual grounding (AgroVG-T1): from an image and a referring expression the model returns the boxes of every matching target, an empty list when none is present, on the AgroVG test split (multi-source crop, weed, fruit and pest imagery); predicted and true boxes matched by maximum-cardinality bipartite matching; higher is better. Source: arxiv.org. Saturation forecast: Around 2030. 13 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1InternVL3.5-8B35.4#1604
2InternVL3.5-38B29.8#1021
3Grounding DINO 1.529.1
4Qwen 3 VL 32B Instruct26.26#606
5DeepSeek-VL2-Small23.44#2464
6Florence-2-Large20.34
7Qwen 3 VL 8B Instruct17.37#1029
8GPT-5.416.53#135
9Gemini 2.5 Pro15.83#276
10OWLv2-Large15.76
11Claude Sonnet 4.611.32#173
12DeepSeek-VL2-Tiny6.16#6534
13GPT-4o (2024-08-06)1.46#743

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=agrovg-multi-target-set-f1 · How It Works · Data refreshed daily, snapshot 2026-11-21.