VisMem: Latent Vision Memory Unlocks Potential of Vision-Language Models

• URL: http://arxiv.org/abs/2511.11007v1
• Date: Fri, 14 Nov 2025 06:51:34 GMT
• Title: VisMem: Latent Vision Memory Unlocks Potential of Vision-Language Models
• Abstract summary: VisMem is a framework that equips Vision-Language Models with dynamic latent vision memories, a short-term module for fine-grained perceptual retention and a long-term module for abstract semantic consolidation.<n>Our experiments reveal that VisMem delivers a significant average performance boost of 11.8% relative to the vanilla model.
• Score: 78.88575188716378
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Despite the remarkable success of Vision-Language Models (VLMs), their performance on a range of complex visual tasks is often hindered by a "visual processing bottleneck": a propensity to lose grounding in visual evidence and exhibit a deficit in contextualized visual experience during prolonged generation. Drawing inspiration from human cognitive memory theory, which distinguishes short-term visually-dominant memory and long-term semantically-dominant memory, we propose VisMem, a cognitively-aligned framework that equips VLMs with dynamic latent vision memories, a short-term module for fine-grained perceptual retention and a long-term module for abstract semantic consolidation. These memories are seamlessly invoked during inference, allowing VLMs to maintain both perceptual fidelity and semantic consistency across thinking and generation. Extensive experiments across diverse visual benchmarks for understanding, reasoning, and generation reveal that VisMem delivers a significant average performance boost of 11.8% relative to the vanilla model and outperforms all counterparts, establishing a new paradigm for latent-space memory enhancement. The code will be available: https://github.com/YU-deep/VisMem.git.

Related papers

• EventVLA: Event-Driven Visual Evidence Memory for Long-Horizon Vision-Language-Action Policies [68.812675280427]
  EventVLA is an end-to-end framework founded on the concept of sparse visual evidence memory.<n>KEM directly predicts future probabilities from the VLA's latent embeddings to autonomously capture and store sparse, task-critical visual events.<n>We propose RoboTwin-MeM, a diagnostic benchmark designed to evaluate non-Markovian manipulation tasks with interactive visual evidence.
  arXiv  Detail & Related papers  (2026-06-18T11:11:37Z)
• M$^3$Eval: Multi-Modal Memory Evaluation through Cognitively-Grounded Video Tasks [19.25978075323521]
  M$3$Eval is the first comprehensive evaluation framework and benchmark for probing different memory dimensions in multi-modal models.<n>We conduct experiments across representative multi-modal models, revealing consistent weaknesses and distinctive behaviors.<n>Our findings highlight memory as a fundamental yet underexplored capability and offer insights for designing more effective memory mechanisms in multi-modal models.
  arXiv  Detail & Related papers  (2026-06-03T15:28:57Z)
• Vision Inference Former: Sustaining Visual Consistency in Multimodal Large Language Models [58.9387276447485]
  Vision Inference Former (VIF) is a lightweight architectural module that establishes a bridge between pure visual representations and the model's output space.<n>We conduct experiments on 14 benchmark tasks covering general reasoning, OCR, table understanding, vision-centric evaluation, and hallucination.
  arXiv  Detail & Related papers  (2026-05-18T10:04:22Z)
• Persistent Visual Memory: Sustaining Perception for Deep Generation in LVLMs [66.15429821530503]
  Persistent Visual Memory is a lightweight learnable module designed to strengthen sustained, on-demand access to visual evidence.<n>Experiments on Qwen3-VL models demonstrate that PVM brings notable improvements with negligible parameter overhead.<n>In-depth analysis reveals that PVM shows improved robustness in longer generations and accelerates internal prediction convergence.
  arXiv  Detail & Related papers  (2026-05-01T17:54:37Z)
• POINTS-Long: Adaptive Dual-Mode Visual Reasoning in MLLMs [64.72502363697518]
  POINTS-Long is a native dual-mode MLLM featuring dynamic visual token scaling inspired by the human visual system.<n>Our work provides new insights into the design of future MLLMs and lays the foundation for adaptive and efficient long-form visual understanding.
  arXiv  Detail & Related papers  (2026-04-13T15:38:22Z)
• RoboMME: Benchmarking and Understanding Memory for Robotic Generalist Policies [54.23445842621374]
  Memory is critical for long-horizon and history-dependent robotic manipulation.<n>Recent vision-language-action (VLA) models have begun to incorporate memory mechanisms.<n>We introduce RoboMME: a large-scale standardized benchmark for evaluating and advancing VLA models.
  arXiv  Detail & Related papers  (2026-03-04T21:59:32Z)
• From Verbatim to Gist: Distilling Pyramidal Multimodal Memory via Semantic Information Bottleneck for Long-Horizon Video Agents [78.30630000529133]
  We propose MM-Mem, a pyramidal multimodal memory architecture grounded in Fuzzy-Trace Theory.<n> MM-Mem memory structures hierarchically into a Sensory Buffer, Episodic Stream, and Symbolic.<n>Experiments confirm the effectiveness of MM-Mem on both offline and streaming tasks.
  arXiv  Detail & Related papers  (2026-03-02T05:12:45Z)
• Toward Cognitive Supersensing in Multimodal Large Language Model [67.15559571626747]
  We introduce Cognitive Supersensing, a training paradigm that endows MLLMs with human-like visual imagery capabilities.<n>In experiments, MLLMs trained with Cognitive Supersensing significantly outperform state-of-the-art baselines on CogSense-Bench.<n>We will open-source the CogSense-Bench and our model weights.
  arXiv  Detail & Related papers  (2026-02-02T02:19:50Z)
• EvolMem: A Cognitive-Driven Benchmark for Multi-Session Dialogue Memory [63.84216832544323]
  EvolMem is a new benchmark for assessing multi-session memory capabilities of large language models (LLMs) and agent systems.<n>To construct the benchmark, we introduce a hybrid data synthesis framework that consists of topic-initiated generation and narrative-inspired transformations.<n>Extensive evaluation reveals that no LLM consistently outperforms others across all memory dimensions.
  arXiv  Detail & Related papers  (2026-01-07T03:14:42Z)
• MemVerse: Multimodal Memory for Lifelong Learning Agents [35.218549149012844]
  We introduce MemVerse, a model-agnostic, plug-and-play memory framework.<n>MemVerse bridges fast parametric recall with hierarchical retrieval-based memory.<n>It enables scalable and adaptive multimodal intelligence.
  arXiv  Detail & Related papers  (2025-12-03T10:06:14Z)
• Vision-Language Memory for Spatial Reasoning [4.486751990718678]
  We present VLM$2$, a vision-language model with persistent memory for spatial reasoning.<n>We show that VLM$2$ achieves state-of-the-art performance among video-only models.
  arXiv  Detail & Related papers  (2025-11-25T18:59:02Z)
• Cross-Layer Vision Smoothing: Enhancing Visual Understanding via Sustained Focus on Key Objects in Large Vision-Language Models [13.17978215666921]
  Large Vision-Language Models (LVLMs) can accurately locate key objects in images, yet their attention to these objects tends to be very brief.<n>Motivated by the hypothesis that sustained focus on key objects can improve LVLMs' visual capabilities, we propose Cross-Layer Vision Smoothing (CLVS)<n> CLVS achieves state-of-the-art performance on a variety of visual understanding tasks.
  arXiv  Detail & Related papers  (2025-09-16T09:54:01Z)
• FindingDory: A Benchmark to Evaluate Memory in Embodied Agents [49.18498389833308]
  We introduce a new benchmark for long-range embodied tasks in the Habitat simulator.<n>This benchmark evaluates memory-based capabilities across 60 tasks requiring sustained engagement and contextual awareness.
  arXiv  Detail & Related papers  (2025-06-18T17:06:28Z)
• Cracking the Code of Hallucination in LVLMs with Vision-aware Head Divergence [69.86946427928511]
  We investigate the internal mechanisms driving hallucination in large vision-language models (LVLMs)<n>We introduce Vision-aware Head Divergence (VHD), a metric that quantifies the sensitivity of attention head outputs to visual context.<n>We propose Vision-aware Head Reinforcement (VHR), a training-free approach to mitigate hallucination by enhancing the role of vision-aware attention heads.
  arXiv  Detail & Related papers  (2024-12-18T15:29:30Z)
• Stable Hadamard Memory: Revitalizing Memory-Augmented Agents for Reinforcement Learning [64.93848182403116]
  Current deep-learning memory models struggle in reinforcement learning environments that are partially observable and long-term. We introduce the Stable Hadamard Memory, a novel memory model for reinforcement learning agents. Our approach significantly outperforms state-of-the-art memory-based methods on challenging partially observable benchmarks.
  arXiv  Detail & Related papers  (2024-10-14T03:50:17Z)
• Look Twice Before You Answer: Memory-Space Visual Retracing for Hallucination Mitigation in Multimodal Large Language Models [26.32657568461926]
  multimodal large language models (MLLMs) are prone to hallucinations.<n>MemVR is a novel decoding paradigm inspired by common cognition.<n>MemVR significantly mitigates hallucination across various MLLMs.
  arXiv  Detail & Related papers  (2024-10-04T16:30:54Z)
• Vision Mamba: Efficient Visual Representation Learning with Bidirectional State Space Model [48.233300343211205]
  We propose a new generic vision backbone with bidirectional Mamba blocks (Vim) Vim marks the image sequences with position embeddings and compresses the visual representation with bidirectional state space models. The results demonstrate that Vim is capable of overcoming the computation & memory constraints on performing Transformer-style understanding for high-resolution images.
  arXiv  Detail & Related papers  (2024-01-17T18:56:18Z)

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.