Research map / 3D vision / Human pose estimation
Gaze Target Estimation Anywhere with Concepts
Xu Cao, Houze Yang, Vipin Gunda, Zhongyi Zhou, Tianyu Xu, Adarsh Kowdle, Inki Kim, James M. Rehg
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add text/visual subject prompt (ablation) | GazeAnywhere-DINOv3-L · GazeFollow-Concept | Avg L2: 0.144 -> 0.099 | Improved |
| use full text prompt (all four categories) instead of appearance only (ablation) | GazeAnywhere-DINOv3-L · GazeFollow-Concept | Avg L2: 0.113 -> 0.099 | Improved |
| add head box loss (ablation) | GazeAnywhere-DINOv3-L · VAT-Concept | L2: 0.136 -> 0.123 | Improved |
| add presence loss on top of head loss (ablation) | GazeAnywhere-DINOv3-L · GazeFollow-Concept | Avg L2: 0.099 -> 0.099 | No clear effect |
| add presence loss (ablation) | GazeAnywhere-DINOv3-L · VAT-Concept | AP: 0.863 -> 0.879 | Improved |
| keep image and text encoders frozen (ablation) | GazeAnywhere-DINOv3-L · GazeFollow-Concept | Avg L2: 0.150 -> 0.099 | Improved |
| replace CLIP-B with DINOv3-L encoder (ablation) | GazeAnywhere · GazeFollow-Concept | Avg L2: 0.123 -> 0.099 | Improved |
| replace MetaCLIP2-H (1.9B) with DINOv3-L (870M) (ablation) | GazeAnywhere · VAT-Concept | L2: 0.150 -> 0.123 | Improved |
| increase detector dimension D from 64 to 128 (ablation) | GazeAnywhere-DINOv3-L · GazeFollow-Concept | Avg L2: 0.115 -> 0.100 | Improved |
| increase detector dimension D from 256 to 512 (ablation) | GazeAnywhere-DINOv3-L · VAT-Concept | AUC: 0.928 -> 0.917 | No clear effect |
| increase detector layers from 1 to 3 (ablation) | GazeAnywhere-DINOv3-L · GazeFollow-Concept | Avg L2: 0.126 -> 0.099 | Improved |
| increase detector layers from 3 to 5 (ablation) | GazeAnywhere-DINOv3-L · GazeFollow-Concept | Avg L2: 0.099 -> 0.095 | No clear effect |
Closest papers by meaning
- Gaze-LLE: Gaze Target Estimation via Large-Scale Learned Encoders (CVPR 2025 · similarity 0.90)
- Toward Semantic Gaze Target Detection (NeurIPS 2024 · similarity 0.88)
- Test-Time Personalization with Meta Prompt for Gaze Estimation (AAAI 2024 · similarity 0.87)
- CLIP-Gaze: Towards General Gaze Estimation via Visual-Linguistic Model (AAAI 2024 · similarity 0.87)
- Eyes on VLM: Benchmarking Gaze Following and Social Gaze Prediction in Vision Language Models (NeurIPS 2026 · similarity 0.86)
- TransGOP: Transformer-Based Gaze Object Prediction (AAAI 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
