Upstream training splits; known explicitly identified test/eval rows excluded. Cross-dataset benchmark overlap is not guaranteed. Published as a raw, manually gated release; source annotation caveats remain.

Task views reuse original image archives. Coordinates are normalized 0–1000. Within each query targets are sorted left-to-right then top-to-bottom.

HF preview configs contain 10 examples per view, not the complete training split. Full training uses metadataset.yaml and examples/read_energon.py (Energon 7.4).

Source	View	Records	Targets	Examples
aria_ui	boxes	479,957	20,710,841	10 annotated examples
autogui_704k	boxes	644,383	644,383	10 annotated examples
autogui_704k	points	644,383	644,383	10 annotated examples
jedi	boxes	1,321,383	2,625,055	10 annotated examples
jedi	points	414,821	1,448,370	10 annotated examples
mind2web	boxes	6,891	6,891	10 annotated examples
showui_desktop	boxes	7,496	7,496	10 annotated examples
showui_desktop	points	7,496	7,496	10 annotated examples
showui_web	boxes	21,988	598,856	10 annotated examples
showui_web	points	21,988	598,856	10 annotated examples
uground_8k	points	8,047	79,393	10 annotated examples
uground_coord	points	1,228,158	8,618,494	10 annotated examples
wave_ui	boxes	21,847	21,847	10 annotated examples

What is not claimed

No guarantee of cross-dataset benchmark disjointness, global image uniqueness, semantic annotation correctness, or redistribution/license clearance. Shared box/point views are alternate tasks, not distinct images. Source-specific caveats remain applicable.

The package preserves upstream query text; metadata cleanup does not alter images or embedded image metadata.

Downloads last month: 51

Total file size:

745 GB