Reproduce “Few-shot Personalized Scanpath Prediction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ruoyu Xue, Jingyi Xu, Sounak Mondal, Hieu Le, Gregory Zelinsky, Minh Hoai, Dimitris Samaras
A new method facilitates personalized scanpath prediction for individuals with minimal prior data by employing a decoupled architecture. It achieves significant performance gains over existing techniques while enabling adaptation to new subjects in under four seconds.
Key insights
- Decoupling subject embedding learning from the core scanpath prediction task is crucial for enabling efficient few-shot personalization without costly fine-tuning.
- Combining classification and contrastive losses during SE-Net training effectively fosters discriminative and consistent subject-specific attention trait extraction from complex spatio-temporal gaze data.
- Integrating task embeddings and fixation duration encoding allows the Subject-Embedding Network to capture both top-down (goal-driven) and bottom-up (stimulus-driven) influences on individual attention patterns.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.