realGUI-800K
- Type
- dataset
- Venue
- Hugging Face (community upload by maharshpatelx)
- Year
- 2026
- Source
- huggingface
- Access
- free
- Language
- English
- Added
- 2026-07-17T20:18:03.642398+00:00
- Verified
- 2026-07-17T20:18:03.642398+00:00
Summary
A GUI grounding dataset of ~794,000 image-question-answer triples for training small Vision Language Models to locate GUI elements. It was generated by transforming 80,000 base GUI elements (sourced from the agentsea/wave-ui dataset) into 800,000 question-answer pairs using LLM-powered paraphrase generation, where each answer is a bounding box locating the referenced UI element. It has been used to fine-tune LiquidAI/LFM2-VL-450M for GUI element detection.
Keywords
gui vision vlm grounding bounding-box ui agents paraphrase synthetic
Topics
Vision / GUI Agents
Research notes
- Generated from agentsea/wave-ui base elements via LLM paraphrase augmentation. Code available at GitHub: MaharshPatelX/realGUI-800K.