New benchmark and method advance open-vocabulary gaze object prediction

Researchers have introduced a new benchmark called Diverse Scenes for Gaze object prediction (DiSG) to evaluate Open-Vocabulary Gaze Object Prediction (OVGOP). This task aims to identify and recognize objects that humans are looking at, even if those objects are not part of a predefined set of categories. The proposed framework uses text-driven object discovery and a gaze-guided selection module to find potential targets. Additionally, a technique called Gradient-Informed Selection Tuning (GIST) is used to fine-tune model parameters relevant to specific vocabularies, improving performance in both open- and closed-vocabulary settings. AI

IMPACT Advances the ability of AI systems to understand human attention and interaction in real-world scenarios by enabling gaze prediction for unseen object categories.

RANK_REASON The cluster describes a new academic paper introducing a benchmark and method for a computer vision task. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• Diverse Scenes for Gaze object prediction (DiSG)
• Gradient-Informed Selection Tuning (GIST)
• Open-Vocabulary Gaze Object Prediction (OVGOP)

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →