cua-lite provides a preprocessed version of the GUIAct dataset for GUI action modeling. The data spans web single-step grounding, web multi-step navigation, and Android multi-step navigation tasks, with normalized coordinates. The upstream test split is honored as the validation split.
Use Cases
- Train models for single-step GUI grounding based on web action data.
- Develop agents for multi-step web navigation tasks.
- Build Android automation systems based on smartphone navigation sequences.
- Benchmark GUI understanding models across web and mobile domains.
Strengths
- Coordinates are normalized to a consistent [0, 1000] range, which likely aids model training.
- Data covers three distinct GUI interaction domains: web-single, web-multi, and smartphone.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
Provenance
- Source
- https://huggingface.co/datasets/yiye2023/GUIAct
- Collection Method
- Preprocessed version of an upstream dataset; original gathering method is not specified.
- Time Range
- null
- Freshness
- Last updated 2026-06-18 06:38:44; freshness should be verified.
- Geography
- null