Research map / Vision-language and multimodal / CLIP and zero-shot
Gaze-Language Alignment for Zero-Shot Prediction of Visual Search Targets from Human Gaze Scanpaths
Sounak Mondal, Naveen Sendhilnathan, Ting Zhang, Yue Liu, Michael Proulx, Michael Louis Iuzzolino, Chuan Qin, Tanya R. Jonker
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 |
|---|---|---|---|
| replace CWCL with GLAD training (LLM prompts) (ablation) | GLAM · COCO-Search18 zero-shot | target accuracy: 23.82 -> 30.17 | Improved |
| replace CWCL with GLAD training (category label prompts) (ablation) | GLAM · COCO-Search18 zero-shot | target accuracy: 22.34 -> 30.17 | Improved |
| replace CWCL with GLAD (LLM prompts) (ablation) | GLAM · COCO-Search18 fully supervised | target accuracy: 52.95 -> 58.22 | Improved |
| use LLM-generated prompts instead of category labels under CWCL (ablation) | GLAM + CWCL · COCO-Search18 fully supervised | target accuracy: 57.95 -> 52.95 | Hurt |
| use LLM-generated prompts instead of category labels under CWCL (ablation) | GLAM + CWCL · COCO-Search18 zero-shot | target accuracy: 22.34 -> 23.82 | No clear effect |
| add GLAD with LLM prompts versus classification-based GLAM (zero-shot method vs supervised classifier) | GLAM · COCO-Search18 fully supervised | target accuracy: 46.45 -> 58.19 | Improved |
| add contextual cues to appearance-only LLM prompts (ablation) | GLAM + GLAD · COCO-Search18 fully supervised, all trials | target accuracy: 49.91 -> 58.22 | Improved |
| add appearance cues to context-only LLM prompts (ablation) | GLAM + GLAD · COCO-Search18 fully supervised, target-present | target accuracy: 70.75 -> 77.76 | Improved |
| add appearance cues to context-only LLM prompts (ablation) | GLAM + GLAD · COCO-Search18 fully supervised, target-absent | target accuracy: 37.39 -> 38.69 | No clear effect |
| add peripheral information (w=1 vs w=0) (ablation) | GLAM + GLAD · COCO-Search18 zero-shot | target accuracy: 23.25 -> 30.17 | Improved |
| add peripheral information (w=1 vs w=0) (ablation) | GLAM + GLAD · COCO-Search18 fully supervised | target accuracy: 59.43 -> 58.22 | Hurt |
| use full-image peripheral access instead of window w=1 (ablation) | GLAM + GLAD · COCO-Search18 zero-shot | target accuracy: 30.17 -> 25.79 | Hurt |
Closest papers by meaning
- Eyes on VLM: Benchmarking Gaze Following and Social Gaze Prediction in Vision Language Models (NeurIPS 2026 · similarity 0.88)
- OmniGF: A Dual-Branch Vision-Language Framework for Unified Gaze Following (NeurIPS 2026 · similarity 0.88)
- Voila-A: Aligning Vision-Language Models with User's Gaze Attention (NeurIPS 2024 · similarity 0.88)
- EyEar: Learning Audio Synchronized Human Gaze Trajectory Based on Physics-Informed Dynamics (AAAI 2025 · similarity 0.87)
- Gaze-LLE: Gaze Target Estimation via Large-Scale Learned Encoders (CVPR 2025 · similarity 0.87)
- CLIP-Gaze: Towards General Gaze Estimation via Visual-Linguistic Model (AAAI 2024 · similarity 0.87)
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.
