VistaQA - Evidence Mask mIoU: leaderboard
Metric: Overall evidence mask score (%): mean per-sample mask score (Hungarian-matched IoU, 1 for correctly predicting no mask on a hallucination sample), on the 1,157 VistaQA samples (six task types, six visual domains, 314 hallucination samples whose queried entity is absent), zero-shot with standardized output-format instructions; the answer is judged correct or not by Qwen 2.5-14B and the evidence masks are matched to the reference masks by Hungarian IoU; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 14 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | SAM3 + GPT-5.4 (Thinking) | 41.98 |
| 2 | SAM3 + Qwen3-VL-4B-Instruct | 39.91 |
| 3 | SAM3 + GPT-5.4 | 39.26 |
| 4 | SAM3 + Qwen3-VL-32B-Instruct | 36.02 |
| 5 | SAM3 + Gemini 3 (VistaQA checkpoint unspecified) | 34.99 |
| 6 | R-Sa2VA-Qwen3VL-4B-RL | 24.25 |
| 7 | SESAME-7B | 23.67 |
| 8 | Sa2VA-8B | 21.01 |
| 9 | UniPixel-7B | 20.43 |
| 10 | TreeVGR-7B | 19.66 |
Interactive version: theaggregate.ai/benchmark?slug=vistaqa-evidence-mask-miou · How It Works · Data refreshed daily, snapshot 2026-10-07.