PreView improves GUI grounding without retraining the model
Original title: PreView: Enhancing GUI Grounding with Preprocessed Views and Cues
A lightweight preprocessor adds zoomed views and cues, lifting accuracy for three frozen vision-language models on five benchmarks.
PreView is an auxiliary preprocessor for GUI grounding that prepares inputs for an existing model rather than retraining it. A lightweight model predicts and refines a center, crops a local view, and enlarges it, then adds a red-circle marker and a textual cue about the candidate. The frozen grounding model uses those inputs to predict the click. Across five benchmarks it improved average accuracy for Qwen3-VL-2B, OS-Atlas-4B, and Qwen3-VL-4B.
- PreView does not further train the grounding models it feeds.
- Prepared inputs include a zoomed local view and a red-circle cue.
- Analyses cover views, cues, center refinement, and input budgets.
- Products
- PreView
- AI models
- Qwen3-VL-2BOS-Atlas-4BQwen3-VL-4B
From the original article
1 min read at huggingface.co
GUI grounding requires locating small controls in cluttered screenshots and identifying which control a natural-language instruction refers to. Cropping and enlargement can expose local detail, but the resulting image may still leave ambiguity about the intended target. We propose PreView, an auxiliary preprocessor that improves the inputs to an existing grounding model without further training that model.
By Wenjun Hou, Kaishuai Xu · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.