Searcharxiv⌕ Search

arXiv subjects

Ziyu Guo

Publications and source records attributed to Ziyu Guo.

At least 19 recordsLinked to original sources
arXiv2026

VBVR-Pro: A Scalable and Verifiable Suite for Native Visual Reasoning

Native visual reasoning treats visual generation as the medium of reasoning itself: visual states (i.e. images and videos) are not merely inputs to be understood or outputs to be rendered, but first-class substrates for problem solving beyond language. Yet progress remains bottlenecked by the lack of scalable training tasks, reliable feedback, and controlled comparisons across generative substrates. In this work, we introduce VBVR-Pro, a closed-loop testbed that makes native visual reasoning through generation trainable, verifiable, optimizable, and experimentally controllable. 1) Task scaling. VBVR-Pro turns visual reasoning into a controlled task space of 300 procedurally generated tasks. Models trained on VBVR-Pro show strong transfer beyond the proposed suite across seven external visual reasoning benchmarks such as RISE-Video, MME-CoF-Pro, and BabyVision. 2) Verifiable rewards. VBVR-Pro provides verifiable reward scorers for task-grounded evaluation. Through a systematic study of leading MLLMs as judges, we identify recurring failure modes of the prevalent VLM-as-a-judge paradigm. In contrast, the proposed scorers are grounded in deterministic, task-specific rules, achieve fine-grained alignment with human judgments. Importantly, they serve as reliable reward signals for large-scale multi-task reinforcement learning and demonstrate stronger post-RL performance across visual reasoning tasks. 3) Mechanism study. VBVR-Pro enables controlled modality studies across more than 30 image, video, and interleaved generators. Our analysis shows that video generation remains strongest for tasks requiring persistent spatiotemporal state tracking, while interleaved generation provides a compute-efficient alternative. Critically, ablations and probing suggest the presence of vision-native trajectories that are crucial to visual reasoning. We release all data, models, scorers, and code.

cs.CV↗
arXiv2026

Transient Chirp Dynamics in Terahertz Quantum Cascade Lasers

Laser frequency chirp is a ubiquitous dynamical process in semiconductor lasers, vital for frequency-modulated photonic systems. In the mid-infrared (MIR) and terahertz (THz) ranges, quantum cascade lasers (QCLs) are ideal sources with high power, narrow linewidth and compact size. While chirp dynamics in MIR QCLs have been studied, the transient chirp behavior of THz QCLs--particularly the thermal chirp on microsecond to millisecond timescales--remains largely unexplored. Here, we experimentally investigate transient thermal chirp dynamics in single-mode THz QCLs via an on-chip heterodyne scheme. Twin monolithically integrated single-mode QCLs are used: one pulsed QCL as the device under test, and one continuous-wave (CW) QCL serving as both local oscillator (LO) and ultrafast THz detector. The frequency chirp is mapped to the radio-frequency (RF) domain by heterodyne down-conversion. By varying current and temperature, we observe three distinct chirp features: unidirectional down-chirp, V-shaped chirp, and unidirectional up-chirp. A two-node thermal model reproduces the dynamics with good agreement with experiments. Chirp dynamics in the multi-mode regime are also identified, showing the potential for sensitive dynamic spectral characterization. These findings deepen the understanding of THz QCL thermal chirp mechanisms and support applications in THz frequency combs, frequency-modulated continuous-wave (FMCW) radar, and high-speed coherent communications.

physics.optics↗
arXiv2026

VideoCoCo: Code-as-CoT for Physically-Consistent Video Generation via an Agentic Dual-Engine System

Text-to-video models have achieved remarkable visual quality, yet they still struggle to generate physically consistent dynamics because the temporal evolution of a scene must be inferred implicitly from a highly compressed text prompt. Existing chain-of-thought approaches introduce intermediate plans or visual states, but these representations are typically non-executable or temporally sparse, limiting their ability to instantiate and control the complete spatiotemporal process. To address this limitation, we introduce VideoCoCo, an agentic dual-engine framework in which executable Blender code serves as a process-level chain of thought. Given a text prompt, a coding agent synthesizes a Blender program that explicitly specifies the scene and its temporal evolution. The executable simulation engine runs the program to produce a deterministic spatiotemporal draft, which is subsequently transformed into a photorealistic video by a generative video engine through draft-conditioned editing. This decomposition separates process-level reasoning from high-fidelity visual realization. To adapt the video editor to simulated drafts, we construct VideoCoCo-3K, a curated dataset of draft-instruction-target triplets. VideoCoCo improves the OmniWeaving baseline from 0.475 to 0.558 on PhyGenBench and from 52.18 to 77.88 on VBench-2.0, achieving the best average score on both benchmarks. These results demonstrate that executable code provides an effective, controllable, and inspectable intermediate representation for physically consistent video generation.

cs.CV↗
arXiv2026

OpenCoF: Learning to Reason Through Video Generation

Reasoning has become a core capability for large models, especially when reliable decisions require understanding logical consequences. Recent video generation models offer a reasoning path distinct from previous Chain-of-Thought (CoT): reasoning can unfold through temporally connected frames, known as Chain-of-Frame (CoF) reasoning. However, existing video generators are primarily trained on general video corpora, still lacking diverse supervision and dedicated designs for CoF reasoning. To address this gap, we introduce OpenCoF, a framework comprising the OpenCoF-17K dataset, a reasoning video dataset spanning 11 task families, and Wan-CoF, a fine-tuned video model for studying whether diverse temporal supervision improves CoF behavior. Across four video reasoning benchmarks, Wan-CoF achieves considerable gains over the Wan2.2-I2V-A14B baseline. Building on this, we empirically explore more advanced designs for CoF capabilities, i.e., equipping the model with visual and textual reasoning tokens. This mechanism respectively captures low-level visual cues and high-level semantic priors for spatial and temporal reasoning. Through performance comparisons and attention analysis, we examine how these tokens contribute across model depth, denoising steps, space, and time. Our results suggest that stronger video reasoning requires both broad temporal supervision and explicit mechanisms for organizing intermediate reasoning state. We open-source the dataset, model, and code to facilitate future research on reasoning-oriented video generation.

cs.CV↗
arXiv2026

LaST$_{0}$: Latent Spatio-Temporal Chain-of-Thought for Robotic Vision-Language-Action Model

Vision-Language-Action (VLA) models have recently shown strong generalization, with some approaches seeking to explicitly generate linguistic reasoning traces or predict future observations prior to execution. However, explicit reasoning typically incurs non-negligible inference latency, which constrains the temporal resolution required for robotic manipulation. Moreover, such reasoning is confined to the linguistic space, imposing a representational bottleneck that struggles to faithfully capture ineffable physical attributes. To mitigate these limitations, we propose LaST$_0$, a framework that enables efficient reasoning before acting through a Latent Spatio-Temporal Chain-of-Thought (CoT), capturing fine-grained physical and robotic dynamics that are often difficult to verbalize. Specifically, we introduce a token-efficient latent CoT space that models future visual dynamics, 3D structural information, and robot proprioceptive states, and further extends these representations across time to enable temporally consistent implicit reasoning trajectories. Furthermore, LaST$_0$ adopts a dual-system architecture implemented via a Mixture-of-Transformers design, where a reasoning expert conducts low-frequency latent inference and an acting expert generates high-frequency actions conditioned on robotics-oriented latent representations. To facilitate coordination, LaST$_0$ is trained with heterogeneous operation frequencies, enabling adaptive switching during deployment. Across 10 real-world tasks spanning tabletop, mobile, and dexterous hand manipulation, LaST$_0$ improves mean success rates by 13%, 14% and 14% over prior SOTA VLA methods, respectively.

cs.RO↗
arXiv2026

New Dynamical Measurements from a Lensed Quasar Sample: Joint Analysis Constrains the Mass Profile Evolution of Lens Galaxies

We present a systematic study of the internal mass structure of early-type galaxies (ETGs) based on 106 galaxy-scale strong gravitational lenses with background quasars, all having spectroscopic redshifts. From this parent sample, we select 24 systems with high-quality ancillary data for joint analysis of strong lensing geometry and stellar kinematics. A key contribution is the derivation of new single-aperture stellar velocity dispersions for 11 lens galaxies via an iterative spectroscopic fitting procedure that mitigates quasar contamination, providing previously unavailable data. We model the total mass-density profile as a power law, $ρ\propto r^{-γ}$, and parameterise its logarithmic slope as $γ= γ_0 + γ_z \cdot z_l + γ_s \cdot \log \tildeΣ$, where $z_l$ is the lens redshift and $\tildeΣ$ the surface mass density. Within a flat $Λ$CDM framework and using DESI BAO measurements as a prior, we constrain the parameters via Monte Carlo nested sampling to $γ_0 = 1.62^{+0.11}_{-0.12}$, $γ_z = -0.35^{+0.08}_{-0.09}$, and $γ_s = 0.37^{+0.08}_{-0.07}$ ($68\%$ confidence intervals). Our results robustly demonstrate that $γ$ increases with surface mass density ($γ_s > 0$) and decreases with redshift ($γ_z < 0$). This implies that, at fixed redshift, galaxies with denser stellar cores have steeper mass profiles, while at fixed density, profiles become shallower at higher redshifts. By successfully applying the joint lensing--dynamics method to a substantial, independently acquired sample of lensed quasars, this work provides crucial validation of structural trends previously observed in galaxy--galaxy lensing systems, reinforcing the established evolutionary picture for massive ETGs and establishing lensed quasars as a potent probe of galaxy structure.

astro-ph.GA↗
arXiv2026

VGGT-Edit: Feed-forward Native 3D Scene Editing with Residual Field Prediction

High-quality 3D scene reconstruction has recently advanced toward generalizable feed-forward architectures, enabling the generation of complex environments in a single forward pass. However, despite their strong performance in static scene perception, these models remain limited in responding to dynamic human instructions, which restricts their use in interactive applications. Existing editing methods typically rely on a 2D-lifting strategy, where individual views are edited independently and then lifted back into 3D space. This indirect pipeline often leads to blurry textures and inconsistent geometry, as 2D editors lack the spatial awareness required to preserve structure across viewpoints. To address these limitations, we propose VGGT-Edit, a feed-forward framework for text-conditioned native 3D scene editing. VGGT-Edit introduces depth-synchronized text injection to align semantic guidance with the backbone's spatial poses, ensuring stable instruction grounding. This semantic signal is then processed by a residual transformation head, which directly predicts 3D geometric displacements to deform the scene while preserving background stability. To ensure high-fidelity results, we supervise the framework with a multi-term objective function that enforces geometric accuracy and cross-view consistency. We also construct the DeltaScene Dataset, a large-scale dataset generated through an automated pipeline with 3D agreement filtering to ensure ground-truth quality. Experiments show that VGGT-Edit substantially outperforms 2D-lifting baselines, producing sharper object details, stronger multi-view consistency, and near-instant inference speed. The project page is https://chriszkxxx.github.io/VGGT-Edit/.

cs.CV↗
arXiv2026

ATLAS: Agentic or Latent Visual Reasoning? One Word is Enough for Both

Visual reasoning, often interleaved with intermediate visual states, has emerged as a promising direction in the field. A straightforward approach is to directly generate images via unified models during reasoning, but this is computationally expensive and architecturally non-trivial. Recent alternatives include agentic reasoning through code or tool calls, and latent reasoning with learnable hidden embeddings. However, agentic methods incur context-switching latency from external execution, while latent methods lack task generalization and are difficult to train with autoregressive parallelization. To combine their strengths while mitigating their limitations, we propose ATLAS, a framework in which a single discrete 'word', termed as a functional token, serves both as an agentic operation and a latent visual reasoning unit. Each functional token is associated with an internalized visual operation, yet requires no visual supervision and remains a standard token in the tokenizer vocabulary, which can be generated via next-token prediction. This design avoids verbose intermediate visual content generation, while preserving compatibility with the vanilla scalable SFT and RL training, without architectural or methodological modifications. To further address the sparsity of functional tokens during RL, we introduce Latent-Anchored GRPO (LA-GRPO), which stabilizes the training by anchoring functional tokens with a statically weighted auxiliary objective, providing stronger gradient updates. Extensive experiments and analyses demonstrate that ATLAS achieves superior performance on challenging benchmarks while maintaining clear interpretability. We hope ATLAS offers a new paradigm inspiring future visual reasoning research.

cs.CV↗
arXiv2026

AIA: Rethinking Architecture Decoupling Strategy In Unified Multimodal Model

Unified multimodal models for image generation and understanding represent a significant step toward AGI and have attracted widespread attention from researchers. The main challenge of this task lies in the difficulty in establishing an optimal training paradigm due to inherent conflicting targets in understanding and generation tasks. To alleviate these conflicts and pursue higher performance, many researchers adopt varying degrees of architecture decoupling (e.g., Double image encoders, MOE/MOT architecture, or frozen MLLM). However, excessive model decoupling can lead to the loss of interleave generation ability, undermining the original intent of unified models. In this work, we aim to explore how to mitigate task conflicts without resorting to model decoupling. Firstly, we analyze why decoupling boosts performance by studying the cross-modal attention behavior of models. We observe that architecture decoupling does not solve task conflicts, but essentially drives models toward cross-modal interaction patterns of task-specific models, as seen in Qwen3-VL and HunyuanImage-3.0, and that the more thorough the decoupling, the more consistent the behavior becomes. Motivated by this observation, we propose Attention Interaction Alignment (AIA) loss, which explicitly learns task-specific multimodal interaction patterns during training. To demonstrate the generalizability of our AIA loss, we apply it to Emu3 and Janus-Pro during SFT and post-training stage respectively. Without bells and whistles, AIA not only refines cross-modal attention patterns, but also boosts both generation and understanding performance.

cs.CV↗
arXiv2026

Plan in Sandbox, Navigate in Open Worlds: Learning Physics-Grounded Abstracted Experience for Embodied Navigation

Vision-Language Models (VLMs) have demonstrated exceptional general reasoning capabilities. However, their performance in embodied navigation remains hindered by a scarcity of aligned open-world vision and robot control data. Despite simulators providing a cost-effective alternative for data collection, the inherent reliance on photorealistic simulations often limits the transferability of learned policies. To this end, we propose \textit{\textbf{S}andbox-\textbf{A}bstracted \textbf{G}rounded \textbf{E}xperience} (\textbf{\textit{SAGE}}), a framework that enables agents to learn within a physics-grounded semantic abstraction rather than a photorealistic simulation, mimicking the human capacity for mental simulation where plans are rehearsed in simplified physics abstractions before execution. \textit{SAGE} system operates via three synergistic phases: (1) \textit{Genesis}: constructing diverse, physics-constrained semantic environments to bootstrap experience; (2) \textit{Evolution}: distilling experiences through Reinforcement Learning (RL), utilizing a novel asymmetric adaptive clipping mechanism to stabilize updates; (3) \textit{Navigation}: bridging the abstract policy to open-world control. We demonstrate that \textit{SAGE} significantly improves planner-assisted embodied navigation, achieving a 53.21\% LLM-Match Success Rate on A-EQA (+9.7\% over baseline), while showing encouraging transfer to physical indoor robot deployment.

cs.RO↗
arXiv2026

Uni-Synergy: Bridging Understanding and Generation for Personalized Reasoning via Co-operative Reinforcement Learning

Unified Multimodal Models (UMMs) excel in general tasks but struggle to bridge the gap between personalized understanding and generation. Prior works largely rely on implicit token-level alignment via supervised fine-tuning, which fails to fully capture the potential synergy between comprehension and creation. In this work, we propose Sync-R1, an end-to-end reinforcement learning framework that jointly optimizes personalized understanding and generation within a single, explicit reasoning loop. Through this unified feedback process, Sync-R1 enables personalized comprehension to guide content creation, while the resulting generation quality reciprocally refines understanding within an integrated reward landscape. To efficiently orchestrate this dual-task synergy, we introduce Sync-GRPO, a reinforcement learning method utilizing an ensemble reward system. Furthermore, we propose Dynamic Group Scaling (DGS), which adaptively filters low-potential trajectories to reduce gradient variance and accelerate convergence. To better reflect real-world complexity, we introduce UnifyBench++, featuring denser textual descriptions and richer user contexts. Experimental results demonstrate that Sync-R1 achieves state-of-the-art performance, showcasing superior cross-task reasoning and robust personalization without requiring complex cold-start procedures. The code and the UnifyBench++ dataset will be released at: https://github.com/arctanxarc/UniCTokens.

cs.CV↗
arXiv2026

LaST-R1: Reinforcing Robotic Manipulation via Adaptive Physical Latent Reasoning

Robotic foundation models require reasoning over complex visual scenes to execute adaptive actions in dynamic environments. While recent studies on latent-reasoning Vision-Language-Action (VLA) models have demonstrated the capability to capture fine-grained physical dynamics, they remain predominantly confined to static imitation learning, severely limiting their adaptability and generalization. In this paper, we present LaST-R1, a novel reinforcement learning (RL) post-training framework designed to effectively harness "latent reasoning-before-acting" policies. Specifically, we propose Latent-to-Action Policy Optimization (LAPO), a core RL algorithm that jointly optimizes the latent reasoning process and the action generation. By explicitly embedding latent Chain-of-Thought (CoT) reasoning directly within the RL optimization loop, LAPO stimulates profound physical world modeling, which in turn drives robust execution in interactive environments. Furthermore, an adaptive latent CoT mechanism is introduced, allowing the policy to dynamically modulate its reasoning horizon based on diverse environment states. Experiments show that LaST-R1 achieves a near-perfect 99.9% average success rate on the LIBERO benchmark with only one-shot supervised warm-up, significantly improving convergence speed and performance over prior state-of-the-art (SOTA) methods. In real-world deployments, LaST-R1 yields up to a 22.5% average improvement over SOTA supervised fine-tuning approach across four complex tasks, including both single-arm and dual-arm settings. Finally, LaST-R1 demonstrates strong generalization across simulated and real-world environments.

cs.RO↗
arXiv2026

MME-CoF-Pro: Evaluating Reasoning Coherence in Video Generative Models with Text and Visual Hints

Video generative models show emerging reasoning behaviors. It is essential to ensure that generated events remain causally consistent across frames for reliable deployment, a property we define as reasoning coherence. To bridge the gap in literature for missing reasoning coherence evaluation, we propose MME-CoF-Pro, a comprehensive video reasoning benchmark to assess reasoning coherence in video models. Specifically, MME-CoF-Pro contains 303 samples across 16 categories, ranging from visual logical to scientific reasoning. It introduces Reasoning Score as evaluation metric for assessing process-level necessary intermediate reasoning steps, and includes three evaluation settings, (a) no hint (b) text hint and (c) visual hint, enabling a controlled investigation into the underlying mechanisms of reasoning hint guidance. Evaluation results in 7 open and closed-source video models reveals insights including: (1) Video generative models exhibit weak reasoning coherence, decoupled from generation quality. (2) Text hints boost apparent correctness but often cause inconsistency and hallucinated reasoning (3) Visual hints benefit structured perceptual tasks but struggle with fine-grained perception. Website: https://video-reasoning-coherence.github.io/

cs.CV↗
arXiv2026

Quantization Meets dLLMs: A Systematic Study of Post-training Quantization for Diffusion LLMs

Recent advances in diffusion large language models (dLLMs) have introduced a promising alternative to autoregressive (AR) LLMs for natural language generation tasks, leveraging full attention and denoising-based decoding strategies. However, the deployment of these models on edge devices remains challenging due to their massive parameter scale and high resource demands. While post-training quantization (PTQ) has emerged as a widely adopted technique for compressing AR LLMs, its applicability to dLLMs remains largely unexplored. In this work, we present the first systematic study on quantizing diffusion-based language models. We begin by identifying the presence of activation outliers, characterized by abnormally large activation values that dominate the dynamic range. These outliers pose a key challenge to low-bit quantization, as they make it difficult to preserve precision for the majority of values. More importantly, we implement state-of-the-art PTQ methods and conduct a comprehensive evaluation across multiple task types and model variants. Our analysis is structured along four key dimensions: bit-width, quantization method, task category, and model type. Through this multi-perspective evaluation, we offer practical insights into the quantization behavior of dLLMs under different configurations. We hope our findings provide a foundation for future research in efficient dLLM deployment. Our code is publicly available at https://github.com/FelixMessi/QDLM.

cs.CL↗
arXiv2026

High-rate Scalable Entanglement Swapping Between Remote Entanglement Sources on Deployed New York City Fibers

Entanglement swapping between photon pairs generated at physically separated nodes over telecommunication fiber infrastructure is an essential step towards the quantum internet, enabling applications such as quantum repeaters, blind quantum computing, distributed quantum computing, and distributed quantum sensing. However, successful networked entanglement swapping relies on generating indistinguishable pairs of photons and preserving them over deployed fibers. This has limited most previous demonstrations to laboratory settings or relied on sophisticated methods to maintain the necessary indistinguishability. Here, we demonstrate a scalable entanglement swapping experiment using naturally indistinguishable entanglement sources based on warm atomic vapor cells. Without sharing lasers or optical frequency references between nodes, nor the need for pulsing the sources, we achieve a swapping rate of nearly 500 pairs/s while maintaining the CHSH parameter above 2. Additionally, we demonstrate the scalability of our method by maintaining the quality of the entanglement swapping on 17.6-km of deployed fibers in NYC, relying on commercially available SPADs at the spoke nodes, SNSPDs at the hub and standard time-synchronization techniques. Our work paves the way for the practical deployment of large-scale hub-and-spoke quantum networks within cities and data centers.

quant-ph↗
arXiv2026

MME-Emotion: A Holistic Evaluation Benchmark for Emotional Intelligence in Multimodal Large Language Models

Recent advances in multimodal large language models (MLLMs) have catalyzed transformative progress in affective computing, enabling models to exhibit emergent emotional intelligence. Despite substantial methodological progress, current emotional benchmarks remain limited, as it is still unknown: (a) the generalization abilities of MLLMs across distinct scenarios, and (b) their reasoning capabilities to identify the triggering factors behind emotional states. To bridge these gaps, we present \textbf{MME-Emotion}, a systematic benchmark that assesses both emotional understanding and reasoning capabilities of MLLMs, enjoying \textit{scalable capacity}, \textit{diverse settings}, and \textit{unified protocols}. As the largest emotional intelligence benchmark for MLLMs, MME-Emotion contains over 6,000 curated video clips with task-specific questioning-answering (QA) pairs, spanning broad scenarios to formulate eight emotional tasks. It further incorporates a holistic evaluation suite with hybrid metrics for emotion recognition and reasoning, analyzed through a multi-agent system framework. Through a rigorous evaluation of 20 advanced MLLMs, we uncover both their strengths and limitations, yielding several key insights: \ding{182} Current MLLMs exhibit unsatisfactory emotional intelligence, with the best-performing model achieving only $39.3\%$ recognition score and $56.0\%$ Chain-of-Thought (CoT) score on our benchmark. \ding{183} Generalist models (\emph{e.g.}, Gemini-2.5-Pro) derive emotional intelligence from generalized multimodal understanding capabilities, while specialist models (\emph{e.g.}, R1-Omni) can achieve comparable performance through domain-specific post-training adaptation. By introducing MME-Emotion, we hope that it can serve as a foundation for advancing MLLMs' emotional intelligence in the future.

cs.CV↗
arXiv2026

GENIUS: Generative Fluid Intelligence Evaluation Suite

Unified Multimodal Models (UMMs) have shown remarkable progress in visual generation. Yet, existing benchmarks predominantly assess $\textit{Crystallized Intelligence}$, which relies on recalling accumulated knowledge and learned schemas. This focus overlooks $\textit{Generative Fluid Intelligence (GFI)}$: the capacity to induce patterns, reason through constraints, and adapt to novel scenarios on the fly. To rigorously assess this capability, we introduce $\textbf{GENIUS}$ ($\textbf{GEN}$ Fluid $\textbf{I}$ntelligence Eval$\textbf{U}$ation $\textbf{S}$uite). We formalize $\textit{GFI}$ as a synthesis of three primitives. These include $\textit{Inducing Implicit Patterns}$ (e.g., inferring personalized visual preferences), $\textit{Executing Ad-hoc Constraints}$ (e.g., visualizing abstract metaphors), and $\textit{Adapting to Contextual Knowledge}$ (e.g., simulating counter-intuitive physics). Collectively, these primitives challenge models to solve problems grounded entirely in the immediate context. Our systematic evaluation of 12 representative models reveals significant performance deficits in these tasks. Crucially, our diagnostic analysis disentangles these failure modes. It demonstrates that deficits stem from limited context comprehension rather than insufficient intrinsic generative capability. To bridge this gap, we propose a training-free attention intervention strategy. Ultimately, $\textbf{GENIUS}$ establishes a rigorous standard for $\textit{GFI}$, guiding the field beyond knowledge utilization toward dynamic, general-purpose reasoning. Our dataset and code will be released at: $\href{https://github.com/arctanxarc/GENIUS}{https://github.com/arctanxarc/GENIUS}$.

cs.LG↗
arXiv2025

CrossLMM: Decoupling Long Video Sequences from LMMs via Dual Cross-Attention Mechanisms

The advent of Large Multimodal Models (LMMs) has significantly enhanced Large Language Models (LLMs) to process and interpret diverse data modalities (e.g., image and video). However, as input complexity increases, particularly with long video sequences, the number of required tokens has grown significantly, leading to quadratically computational costs. This has made the efficient compression of video tokens in LMMs, while maintaining performance integrity, a pressing research challenge. In this paper, we introduce CrossLMM, decoupling long video sequences from LMMs via a dual cross-attention mechanism, which substantially reduces visual token quantity with minimal performance degradation. Specifically, we first implement a significant token reduction from pretrained visual encoders through a pooling methodology. Then, within LLM layers, we employ a visual-to-visual cross-attention mechanism, wherein the pooled visual tokens function as queries against the original visual token set. This module enables more efficient token utilization while retaining fine-grained informational fidelity. In addition, we introduce a text-to-visual cross-attention mechanism, for which the text tokens are enhanced through interaction with the original visual tokens, enriching the visual comprehension of the text tokens. Comprehensive empirical evaluation demonstrates that our approach achieves comparable or superior performance across diverse video-based LMM benchmarks, despite utilizing substantially fewer computational resources.

cs.CV↗