VQA-G Annotator: A Self-Improving Agentic Framework for High-Fidelity Grounded Dataset Synthesis
Tsinghua Science and Technology · 2026-07
DOI: 10.26599/tst.2026.9010076 ↗Kurz gefasst
Fachartikel in der Zeitschrift Tsinghua Science and Technology, 2026-07. Verfasst von Rongsheng Hu und 10 weiteren Autor:innen.
Abstract (Originalsprache)
As artificial intelligence systems increasingly rely on multi-source perception and cross-modal learning for human-like visual understanding, their outputs should be both semantically valid and spatially grounded. However, large-scale vision-language benchmarks with precise grounding remain costly to construct, while existing synthetic pipelines often suffer from distribution drift, error propagation, and weak spatial verification. We introduce VQA-G Annotator, a self-improving agentic framework that combines distribution-aware planning, multi-agent orchestration, and spatial reasoning verification in a closed-loop generation, evaluation, and refinement process. Experiments across four benchmark sources show that VQA-G Annotator achieves an average VQAScore of 0.88 for semantic alignment and an Acc@0.5 of 0.57 for spatial grounding, comparing favorably with strong automatic baselines. Downstream and human evaluations further support the utility of the synthesized data 18 for scalable and trustworthy grounded visual understanding.
Zitieren
APA
Hu, R., Liu, Y., Guan, R., Di, Y., Bao, J., Jin, Y., Shi, H., Hong, Y., Yu, L., Ma, R., & Buyya, R. (2026). VQA-G Annotator: A Self-Improving Agentic Framework for High-Fidelity Grounded Dataset Synthesis. Tsinghua Science and Technology. https://doi.org/10.26599/tst.2026.9010076
Vancouver
Hu R, Liu Y, Guan R, Di Y, Bao J, Jin Y, et al. VQA-G Annotator: A Self-Improving Agentic Framework for High-Fidelity Grounded Dataset Synthesis. Tsinghua Science and Technology. 2026. doi:10.26599/tst.2026.9010076
BibTeX
@article{Hu2026,
author = {Hu, Rongsheng and Liu, Yuan and Guan, Runwei and Di, Yicheng and Bao, Jiayu and Jin, Yi and Shi, Hongjian and Hong, Yuan and Yu, Lin and Ma, Ruhui and Buyya, Rajkumar},
title = {VQA-G Annotator: A Self-Improving Agentic Framework for High-Fidelity Grounded Dataset Synthesis},
journal = {Tsinghua Science and Technology},
year = {2026},
doi = {10.26599/tst.2026.9010076}
}
EUVIMED