Matched Outcomes, Divergent Gaze: How Foveated MLLMs Search Compared to Humans

• URL: http://arxiv.org/abs/2608.16514v1
• Date: Mon, 17 Aug 2026 12:54:39 GMT
• Title: Matched Outcomes, Divergent Gaze: How Foveated MLLMs Search Compared to Humans
• Abstract summary: We compare three general-purpose large language models (MLLMs) with human eye-movement scanpaths on goal-directed search (COCO-Search18)<n>All three share one signature: low-entropy, large-amplitude, self-consistent scanpaths that agree with themselves far more closely than two humans agree with each other.<n>The gap sits on a process axis that answer-alignment and saliency metrics do not measure. Because they miss it, such metrics cannot certify human-like vision, and zero-shot models suit outcome and spatial questions but not temporal, process-level
• Score: 50.71222402701153
• License: http://creativecommons.org/licenses/by-nc-nd/4.0/
• Abstract: Human visual search is serial: the fovea must land on a candidate to confirm it, and those landings form a scanpath. Whether multimodal large language models (MLLMs), given the same foveated input, search as humans do bears on their use as models of human vision and on attention-alignment scores. We compare three general-purpose MLLMs with human eye-movement scanpaths on goal-directed search (COCO-Search18), driving each model fixation by fixation through an identical, human-matched foveated view and assessing it along three axes: the decision of target presence, the efficiency of reaching the target, and the gaze process itself. The axes dissociate. On the decision and on target acquisition the models match or exceed humans, detecting present targets near ceiling and reaching them on the first saccade more often than people do. The gaze process is not human. Under the human-matched condition, all three share one signature: low-entropy, large-amplitude, self-consistent scanpaths that agree with themselves far more closely than two humans agree with each other. That is consistent with a single-pass, non-serial architecture rather than a limit of acuity. Matched retinal input reproduces where humans look but not how the looking unfolds in time, and no degradation regime recovers human-like search at human-like success. The gap sits on a process axis that answer-alignment and saliency metrics do not measure. Because they miss it, such metrics cannot certify human-like vision, and zero-shot models suit outcome and spatial questions but not temporal, process-level ones.

Related papers

• Do vision-language models search like humans? Reasoning tokens as a reaction-time analog in classic visual-search paradigms [0.0]
  I ask whether vision-language models (VLMs) exhibit the same behavioral signatures.<n>I adapt four classic paradigms: feature versus conjunction search, spatial-configuration (T-vs-L) search, enumeration, and the tilted/vertical search asymmetry.
  arXiv  Detail & Related papers  (2026-06-23T18:19:16Z)
• Language Model Goal Selection Differs from Humans' in an Open-Ended Task [2.012425476229879]
  Large language models (LLMs) are increasingly integrated into human decision-making.<n>We assess the validity of LLMs as proxies for human goal selection in a controlled, open-ended learning task borrowed from cognitive science.
  arXiv  Detail & Related papers  (2026-02-06T15:39:54Z)
• HumanPCR: Probing MLLM Capabilities in Diverse Human-Centric Scenes [72.26829188852139]
  HumanPCR is an evaluation suite for probing MLLMs' capacity about human-related visual contexts.<n>Human-P, HumanThought-C, and Human-R feature over 6,000 human-verified multiple choice questions.<n>Human-R offers a challenging manually curated video reasoning test.
  arXiv  Detail & Related papers  (2025-08-19T09:52:04Z)
• Human Cognitive Benchmarks Reveal Foundational Visual Gaps in MLLMs [65.93003087656754]
  VisFactor is a benchmark that digitizes 20 vision-centric subtests from a well-established cognitive psychology assessment.<n>We evaluate 20 frontier Multimodal Large Language Models (MLLMs) from GPT, Gemini, Claude, LLaMA, Qwen, and SEED families.<n>The best-performing model achieves a score of only 25.19 out of 100, with consistent failures on tasks such as mental rotation, spatial relation inference, and figure-ground discrimination.
  arXiv  Detail & Related papers  (2025-02-23T04:21:32Z)
• Evaluating Multiview Object Consistency in Humans and Image Models [68.36073530804296]
  We leverage an experimental design from the cognitive sciences which requires zero-shot visual inferences about object shape. We collect 35K trials of behavioral data from over 500 participants. We then evaluate the performance of common vision models.
  arXiv  Detail & Related papers  (2024-09-09T17:59:13Z)
• AiOS: All-in-One-Stage Expressive Human Pose and Shape Estimation [55.179287851188036]
  We introduce a novel all-in-one-stage framework, AiOS, for expressive human pose and shape recovery without an additional human detection step. We first employ a human token to probe a human location in the image and encode global features for each instance. Then, we introduce a joint-related token to probe the human joint in the image and encoder a fine-grained local feature.
  arXiv  Detail & Related papers  (2024-03-26T17:59:23Z)
• Learning Visibility for Robust Dense Human Body Estimation [78.37389398573882]
  Estimating 3D human pose and shape from 2D images is a crucial yet challenging task. We learn dense human body estimation that is robust to partial observations. We obtain pseudo ground-truths of visibility labels from dense UV correspondences and train a neural network to predict visibility along with 3D coordinates.
  arXiv  Detail & Related papers  (2022-08-23T00:01:05Z)
• Dual networks based 3D Multi-Person Pose Estimation from Monocular Video [42.01876518017639]
  Multi-person 3D pose estimation is more challenging than single pose estimation. Existing top-down and bottom-up approaches to pose estimation suffer from detection errors. We propose the integration of top-down and bottom-up approaches to exploit their strengths.
  arXiv  Detail & Related papers  (2022-05-02T08:53:38Z)
• Partial success in closing the gap between human and machine vision [30.78663978510427]
  A few years ago, the first CNN surpassed human performance on ImageNet. Here we ask: Are we making progress in closing the gap between human and machine vision? We tested human observers on a broad range of out-of-distribution (OOD) datasets.
  arXiv  Detail & Related papers  (2021-06-14T13:23:35Z)
• Monocular 3D Multi-Person Pose Estimation by Integrating Top-Down and Bottom-Up Networks [33.974241749058585]
  Multi-person pose estimation can cause human detection to be erroneous and human-joints grouping to be unreliable. Existing top-down methods rely on human detection and thus suffer from these problems. We propose the integration of top-down and bottom-up approaches to exploit their strengths.
  arXiv  Detail & Related papers  (2021-04-05T07:05:21Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.