Research map / Vision-language and multimodal / Human and embodied modeling
Modeling Human Gaze Behavior with Diffusion Models for Unified Scanpath Prediction
Giuseppe Cartella, Vittorio Cuculo, Alessandro D'Amelio, Marcella Cornia, Giuseppe Boccignone, Rita Cucchiara
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 CLIP text encoder with RoBERTa (ablation) | ScanDiff · COCO-FreeView and COCO-Search18 | MM/SM/SS: MM 0.078 -> 0.110 (COCO-FreeView) | Hurt |
| use DINOv2 instead of ResNet-50 or CLIP visual backbone (ablation) | ScanDiff · COCO-Search18 target-present | MM: CLIP 0.090 / RN50 0.070 -> DINOv2 0.048 | Improved |
| use DINOv2 instead of ResNet-50 visual backbone (ablation) | ScanDiff · COCO-FreeView | MM: RN50 0.049 vs DINOv2 0.078 | Mixed |
| add L_T alignment loss (ablation) | ScanDiff · COCO-Search18 target-present | MM: 0.058 -> 0.048 | Improved |
| add L_T alignment loss (ablation) | ScanDiff · COCO-FreeView | SM/SS: SM 0.011 -> 0.015; MM 0.088 -> 0.078 | Mixed |
| increase diffusion timesteps from 1000 to 1500 (ablation) | ScanDiff · COCO-Search18 target-present | SS: 0.019 -> 0.061 | Hurt |
| reduce diffusion timesteps from 1000 to 200 (ablation) | ScanDiff · COCO-FreeView | MM/SM/SS: MM 0.078 -> 0.043; SM 0.015 -> 0.072 | Mixed |
| condition via cross-attention instead of concatenating noisy gaze with image embedding (ablation) | ScanDiff · COCO-Search18 target-present | MM: 0.439 -> 0.048 | Improved |
| increase maximum scanpath length from 16 to 32 (ablation) | ScanDiff · COCO-FreeView | SM: 0.015 -> 0.204 | Hurt |
| use diffusion-based stochastic sampling to generate multiple scanpaths | ScanDiff · COCO-FreeView | RSS: 0.501 (ChenLSTM-ISP) -> 0.736 | Improved |
Closest papers by meaning
- ST-DiffEye: Diffusion-based Continuous Gaze Generation via Joint Scanpath-Trajectory Modeling (NeurIPS 2026 · similarity 0.91)
- DiffEye: Diffusion-Based Continuous Eye-Tracking Data Generation Conditioned on Natural Images (NeurIPS 2025 · similarity 0.90)
- EyEar: Learning Audio Synchronized Human Gaze Trajectory Based on Physics-Informed Dynamics (AAAI 2025 · similarity 0.88)
- Unifying Top-down and Bottom-up Scanpath Prediction Using Transformers (CVPR 2024 · similarity 0.87)
- GazeFlow: From Human Gaze Behavior to Generative Egocentric Gaze Prediction (NeurIPS 2026 · similarity 0.86)
- Few-shot Personalized Scanpath Prediction (CVPR 2025 · 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.
