DeepGaze3.5-VL treats visual scanpaths as discrete token sequences predicted autoregressively by vision-language models, achieving 2.18 bits IG on MIT1003 and outperforming prior specialized models even with matched encoders.
Title resolution pending
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CV 2years
2026 2verdicts
UNVERDICTED 2roles
background 1polarities
background 1representative citing papers
A mixture model with adaptive KDE and per-image cross-validation raises estimated human fixation consistency by 5-15% median log-likelihood and up to 2 AUC points over fixed-bandwidth Gaussian baselines.
citing papers explorer
-
DeepGaze3.5-VL: Modeling Scanpaths via Autoregressive Token Prediction
DeepGaze3.5-VL treats visual scanpaths as discrete token sequences predicted autoregressively by vision-language models, achieving 2.18 bits IG on MIT1003 and outperforming prior specialized models even with matched encoders.
-
Raising the Ceiling: Better Empirical Fixation Densities for Saliency Benchmarking
A mixture model with adaptive KDE and per-image cross-validation raises estimated human fixation consistency by 5-15% median log-likelihood and up to 2 AUC points over fixed-bandwidth Gaussian baselines.