TextGaze uses LVLM for gaze target estimation · arXiv cs.CV

Researchers have introduced TextGaze, a novel architecture for gaze target estimation that utilizes a Large Vision-Language Model (LVLM) for semantic guidance. This approach aims to overcome the limitations of existing methods, which either require extensive annotations or focus solely on low-level visual saliency. TextGaze extracts visual features and employs an LVLM to generate gaze-aligned textual cues, processed through a transformer-based fusion module with hierarchical text supervision. The model efficiently predicts gaze heatmaps and in-/out-of-frame status, demonstrating competitive performance and robust cross-dataset generalization on four mainstream datasets without additional fine-tuning. AI

IMPACT This research offers a more efficient and generalizable approach to gaze estimation, potentially improving human-computer interaction and accessibility tools.

RANK_REASON The cluster contains an academic paper detailing a new method for gaze target estimation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• arXiv
• Hugging Face
• Large Vision-Language Model
• Transformer++

• paper
• model release

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