arXiv subjects

Shigeru Kitazawa

Publications and source records attributed to Shigeru Kitazawa.

3 recordsLinked to original sources

arXiv2026

Tracing the Arrow of Time: Diagnosing Temporal Information Flow in Video-LLMs

The Arrow-of-Time (AoT) task, determining whether a video plays forward or backward by recognizing temporal irreversibility, is one humans solve with near-perfect accuracy, yet frontier Video Large Language Models (Video-LLMs) perform only modestly above chance. This gap raises a key question: do visual backbones fail to encode temporal information, or does information bottleneck lie elsewhere in the Video-LLM architecture? We address this question by isolating the vision encoder from the Video-LLM and tracing temporal information across the encoder, projector, and LLM. We find that video-centric encoders with explicit temporal modeling encode strong temporal signals, whereas frame-centric encoders do not. However, when video-centric representations are passed through a standard Video-LLM architecture, performance often collapses, revealing a bottleneck of temporal information flow. We identify projector design as a key factor: Q-Former disrupts temporal information, while a time-preserved MLP projection substantially improves the LLM's access to such information. Our layer-wise analysis further shows temporal representation dynamics across encoder layers. Guided by these findings, we build a Video-LLM with temporal-aware video-centric encoder, time-preserved projector, and AoT supervision, surpassing human performance on AoT$_{PPB}$ with 98.1\% accuracy, and improving broader temporal reasoning tasks by up to 6.0 points on VITATECS-Direction and 1.3 points on TVBench. Our results show that temporal reasoning in Video-LLMs requires both effective temporal encoding and reliable transfer of this information to the LLM.

cs.CV↗

arXiv2025

Which Way Does Time Flow? A Psychophysics-Grounded Evaluation for Vision-Language Models

Modern vision-language models (VLMs) excel at many multimodal tasks, yet their grasp of temporal information in video remains weak and has not been adequately evaluated. We probe this gap with a deceptively simple but revealing challenge: judging the arrow of time (AoT)-whether a short clip is played forward or backward. We introduce AoT-PsyPhyBENCH, a psychophysically validated benchmark that tests whether VLMs can infer temporal direction in natural videos using the same stimuli and behavioral baselines established for humans. Our comprehensive evaluation of open-weight and proprietary, reasoning and non-reasoning VLMs reveals that most models perform near chance, and even the best model lags far behind human accuracy on physically irreversible processes (e.g., free fall, diffusion/explosion) and causal manual actions (division/addition) that humans recognize almost instantly. These results highlight a fundamental gap in current multimodal systems: while they capture rich visual-semantic correlations, they lack the inductive biases required for temporal continuity and causal understanding. We release the code and data for AoT-PsyPhyBENCH to encourage further progress in the physical and temporal reasoning capabilities of VLMs.

cs.CV↗

arXiv2024

Emergence of Human-Like Attention in Self-Supervised Vision Transformers: an eye-tracking study

Many models of visual attention have been proposed so far. Traditional bottom-up models, like saliency models, fail to replicate human gaze patterns, and deep gaze prediction models lack biological plausibility due to their reliance on supervised learning. Vision Transformers (ViTs), with their self-attention mechanisms, offer a new approach but often produce dispersed attention patterns if trained with supervised learning. This study explores whether self-supervised DINO (self-DIstillation with NO labels) training enables ViTs to develop attention mechanisms resembling human visual attention. Using video stimuli to capture human gaze dynamics, we found that DINO-trained ViTs closely mimic human attention patterns, while those trained with supervised learning deviate significantly. An analysis of self-attention heads revealed three distinct clusters: one focusing on foreground objects, one on entire objects, and one on the background. DINO-trained ViTs offer insight into how human overt attention and figure-ground separation develop in visual perception.

q-bio.NC↗