Dataset · agentsea
25 thousand annotated graphical interface elements
agentsea/wave-ui-25k
A cleaned sample of screenshots and UI element annotations for training multimodal agents to interact with screens.
What's inside
The dataset includes 25,000 examples of annotated user interface elements (images and text). This is a cleaned subset from a larger collection of ~80,000 examples combined from WebUI, RoboFlow, and GroundUI-18K sources. All data is standardized to a single schema, with duplicates, overlaps, low-quality samples, and extraneous text elements removed.
What you can do with it
- Training UI agents and VLM models to locate buttons and controls on screen
- Testing interface recognition accuracy in web and desktop applications
- Size
- 10K–100K rows
- Downloads this month
- 609
- Likes
- 40
Models trained on it
- Litwein/Qwen3.8-Flash-Next-REAP320-oQ3e-DWQ-MTP-Vision-MTPLXImage text → text 1K
- Litwein/Qwen3.8-Flash-Next-REAP320-oQ3e-DWQ-MTP-Vision-MLXImage text → text 833
- Litwein/Qwen3.8-Flash-Next-REAP320-oQ3e-fp16-DWQ-MTP-Vision-MLXImage text → text 564
- Litwein/Qwen3.8-Flash-Next-REAP320-oQ3e-fp16-DWQ-MTP-Vision-MTPLXImage text → text 430
- miketes/wave-ui-3b 14
- Saivamsim26/qwen2.5-vl-7b-ui-groundingImage text → text 13
- miketes/wave-ui-7b 13
- minhvn4/florence2-wave-ui-loraImage → text 7
How to load it
from datasets import load_dataset
ds = load_dataset("agentsea/wave-ui-25k")