PreView: Enhancing GUI Grounding with Preprocessed Views and Cues
PreView improves GUI click accuracy across five benchmarks while leaving the underlying grounding models frozen.
A trained lightweight auxiliary model locates a candidate control, refines its center, and crops and enlarges the surrounding screenshot. It adds a red-circle marker with a location hint and a text cue describing the candidate’s identity or function. The grounding model uses those inputs to predict the final click. The authors report improved average accuracy for Qwen3-VL-2B, OS-Atlas-4B, and Qwen3-VL-4B; the abstract gives no numerical gains and labels the paper a work in progress.
HF Daily Papers' note
A trained lightweight auxiliary model locates a candidate control, refines its center, and crops and enlarges the surrounding screenshot. It adds a red-circle marker with a location hint and a text cue describing the candidate’s identity or function. The grounding model uses those inputs to predict the final click. The authors report improved average accuracy for Qwen3-VL-2B, OS-Atlas-4B, and Qwen3-VL-4B; the abstract gives no numerical gains and labels the paper a work in progress.
HF Daily Papers' note
score 4