Visual Token Pruning

High-resolution Multimodal Large Language Models (MLLMs) face prohibitive computational costs during inference due to the explosion of visual tokens. Existing acceleration strategies, such as token pruning or layer sparsity, suffer from severe "backbone dependency", performing well on Vicuna or Mistral architectures (e.g., LLaVA) but causing significant performance degradation when transferred to architectures like Qwen. To address this, we leverage truncated matrix entropy to uncover a universal three-stage inference lifecycle, decoupling visual redundancy into universal Intrinsic Visual Redundancy (IVR) and architecture-dependent Secondary Saturation Redundancy (SSR). Guided by this insight, we propose HalfV, a framework that first mitigates IVR via a unified pruning strategy and then adaptively handles SSR based on its specific manifestation. Experiments demonstrate that HalfV achieves superior efficiency-performance trade-offs across diverse backbones. Notably, on Qwen25-VL, it retains 96.8% performance at a 4.1$\times$ FLOPs speedup, significantly outperforming state-of-the-art baselines. Our code is available at https://github.com/civilizwa/HalfV.

Recent advances have explored visual token pruning to accelerate the inference of large vision-language models (LVLMs). However, existing methods often struggle to balance token importance and diversity: importance-based methods tend to retain redundant tokens, whereas diversity-based methods may overlook informative ones. This trade-off becomes especially problematic under high reduction ratios, where preserving only a small subset of visual tokens is critical. To address this issue, we propose ID-Selection, a simple yet effective token selection strategy for efficient LVLM inference. The key idea is to couple importance estimation with diversity-aware iterative selection: each token is first assigned an importance score, after which high-scoring tokens are selected one by one while the scores of similar tokens are progressively suppressed. In this way, ID-Selection preserves informative tokens while reducing redundancy in a unified selection process. Extensive experiments across 5 LVLM backbones and 16 main benchmarks demonstrate that ID-Selection consistently achieves superior performance and efficiency, especially under extreme pruning ratios. For example, on LLaVA-1.5-7B, ID-Selection prunes 97.2% of visual tokens, retaining only 16 tokens, while reducing inference FLOPs by over 97% and preserving 91.8% of the original performance, all without additional training.

Vision-Language-Action (VLA) models integrate visual perception, language understanding, and action decision-making for cross-modal semantic alignment, exhibiting broad application potential. However, the joint processing of high-dimensional visual features, complex linguistic inputs, and continuous action sequences incurs significant computational overhead and low inference efficiency, thereby hindering real-time deployment and reliability. To address this issue, we use image entropy to quantify the grayscale distribution characteristics of each visual token and introduce attention entropy to capture the distribution of attention scores over task-related text. Visual entropy identifies texture-rich or structurally informative regions, while attention entropy pinpoints semantically relevant tokens. Combined with timestep information, these metrics enable a dynamic transition strategy that shifts the model's focus from global visual features to attention-guided local informative regions. Thus, the resulting VLA-InfoEntropy method integrates spatial, semantic, and temporal cues to reduce redundancy while preserving critical content. Extensive experiments show that our method reduces inference parameters, accelerates inference speed, and outperforms existing approaches.

Transformers are strong baselines in both vision and language because self-attention captures long-range dependencies across tokens. However, the cost of self-attention grows quadratically with the number of tokens. Patch pruning mitigates this cost by estimating per-patch importance and removing redundant patches. To identify informative patches for pruning, we introduce a criterion based on the Shannon entropy of the attention distribution. Low-entropy patches, which receive selective and concentrated attention, are kept as important, while high-entropy patches with attention spread across many locations are treated as redundant. We also extend the criterion from Shannon to Rényi entropy, which emphasizes sharp attention peaks and supports pruning strategies that adapt to task needs and computational limits. In experiments on fine-grained image recognition, where patch selection is critical, our method reduced computation while preserving accuracy. Moreover, adjusting the pruning policy through the Rényi entropy measure yields further gains and improves the trade-off between accuracy and computation.

Large Vision-Language Models (LVLMs) suffer from prohibitive inference costs due to the massive number of visual tokens processed by the language decoder. Existing pruning methods often lead to significant performance degradation because the irreversible removal of visual tokens causes a distribution shift in the hidden states that deviates from the pre-trained full-token regime. To address this, we propose Representation Consistency Pruner, which we refer to as RCP, as a novel framework that integrates cumulative visual token pruning with a delayed repair mechanism. Specifically, we introduce a cross-attention pruner that leverages the intrinsic attention of the LLM as a baseline to predict cumulative masks, ensuring consistent and monotonic token reduction across layers. To compensate for the resulting information loss, we design a delayed repair adapter denoted as DRA, which caches the essence of pruned tokens and applies FiLM-based modulation specifically to the answer generation tokens. We employ a repair loss to match the first and second-order statistics of the pruned representations with a full-token teacher. RCP is highly efficient because it trains only lightweight plug-in modules while allowing for physical token discarding at inference. Extensive experiments on LVLM benchmarks demonstrate that RCP removes up to 88.9% of visual tokens and reduces FLOPs by up to 85.7% with only a marginal average accuracy drop, and outperforms prior methods that avoid fine-tuning the original model on several widely used benchmarks.

For multimodal large language models (MLLMs), visual information is relatively sparse compared with text. As a result, research on visual pruning emerges for efficient inference. Current approaches typically measure token importance based on the attention scores in the visual encoder or in the LLM decoder, then select visual tokens with high attention scores while pruning others. In this paper, we pursue a different and more surgical approach. Instead of relying on mechanism-specific signals, we directly compute Mutual Information (MI) between visual and textual features themselves, prior to their interaction. This allows us to explicitly measure crossmodal dependency at the feature levels. Our MI-Pruner is simple, efficient and non-intrusive, requiring no access to internal attention maps or architectural modifications. Experimental results demonstrate that our approach outperforms previous attention-based pruning methods with minimal latency.

Vision-Language Models (VLMs) have advanced rapidly within the unified Transformer architecture, yet their deployment on resource-constrained devices remains challenging due to high computational complexity. While pruning has emerged as an effective technique for compressing VLMs, existing approaches predominantly focus on a single mode by pruning either parameters or tokens, neglecting fully exploring the inherent redundancy in each mode, which leads to substantial performance degradation at high pruning ratios. To address the above limitations, we propose Collaborative Multi-Mode Pruning (CoMP), a novel framework tailored for VLMs by performing joint parameter and token pruning. Specifically, we first design a Collaborative Importance Metric (CIM) that investigates the mutual interference between the coupled parameters and tokens. It incorporates distinct significance of tokens into the computation of parameter importance scores, while simultaneously mitigating the affect of pruned parameters on token importance scores. Moreover, we develop a Multi-Mode Pruning Strategy (MPS) that decomposes the overall pruning process into a sequence of pruning stages, while in each stage we estimate the priory of different pruning modes based on their pruning cost and adaptively shift to the optimal one. Additionally, MPS integrates the historical cost and random exploration, in order to achieve a stable pruning process and avoid local optimum. Extensive experiments across various vision-language tasks and models demonstrate that our method effectively promotes the performance under high pruning ratios by comparing to the state-of-the-art approaches. The source code is available at https://github.com/Wuzimeng/CoMP.git.

Multimodal Large Language Models (MLLMs) have shown strong reasoning ability, but their high computational and memory costs hinder deployment in resource-constrained settings. While Post-Training Quantization (PTQ) and vision token pruning are standard compression techniques, they are usually treated as independent optimizations. In this paper, we show that these two techniques are strongly coupled: naively applying semantic-based token pruning to PTQ-optimized MLLMs can discard activation outliers that are important for numerical stability and thus worsen quantization errors in low-bit regimes (e.g., W4A4). To address this issue, we propose a quantization-aware vision token pruning framework. Our method introduces a lightweight hybrid sensitivity metric that combines simulated group-wise quantization error with outlier intensity. By combining this metric with standard semantic relevance scores, the method retains tokens that are both semantically informative and robust to quantization. Experiments on standard LLaVA architectures show that our method consistently outperforms naive integration baselines. At an aggressive pruning ratio that retains only 12.5% of visual tokens, our framework improves accuracy by 2.24% over the baseline and even surpasses dense quantization without pruning. To the best of our knowledge, this is the first method that explicitly co-optimizes vision token pruning and PTQ for accurate low-bit MLLM inference.

Recent advances in Multimodal Large Language Models (MLLMs) have expanded reasoning capabilities into 3D domains, enabling fine-grained spatial understanding. However, the substantial size of 3D MLLMs and the high dimensionality of input features introduce considerable inference overhead, which limits practical deployment on resource constrained platforms. To overcome this limitation, this paper presents Efficient3D, a unified framework for visual token pruning that accelerates 3D MLLMs while maintaining competitive accuracy. The proposed framework introduces a Debiased Visual Token Importance Estimator (DVTIE) module, which considers the influence of shallow initial layers during attention aggregation, thereby producing more reliable importance predictions for visual tokens. In addition, an Adaptive Token Rebalancing (ATR) strategy is developed to dynamically adjust pruning strength based on scene complexity, preserving semantic completeness and maintaining balanced attention across layers. Together, they enable context-aware token reduction that maintains essential semantics with lower computation. Comprehensive experiments conducted on five representative 3D vision and language benchmarks, including ScanRefer, Multi3DRefer, Scan2Cap, ScanQA, and SQA3D, demonstrate that Efficient3D achieves superior performance compared with unpruned baselines, with a +2.57% CIDEr improvement on the Scan2Cap dataset. Therefore, Efficient3D provides a scalable and effective solution for efficient inference in 3D MLLMs. The code is released at: https://github.com/sol924/Efficient3D

Video temporal grounding (VTG) is a critical task in video understanding and a key capability for extending video large language models (Vid-LLMs) to broader applications. However, existing Vid-LLMs rely on uniform frame sampling to extract video information, resulting in a sparse distribution of key frames and the loss of crucial temporal cues. To address this limitation, we propose Grounded Visual Token Sampling (GroundVTS), a Vid-LLM architecture that focuses on the most informative temporal segments. GroundVTS employs a fine-grained, query-guided mechanism to filter visual tokens before feeding them into the LLM, thereby preserving essential spatio-temporal information and maintaining temporal coherence. Futhermore, we introduce a progressive optimization strategy that enables the LLM to effectively adapt to the non-uniform distribution of visual features, enhancing its ability to model temporal dependencies and achieve precise video localization. We comprehensively evaluate GroundVTS on three standard VTG benchmarks, where it outperforms existing methods, achieving a 7.7-point improvement in mIoU for moment retrieval and 12.0-point improvement in mAP for highlight detection. Code is available at https://github.com/Florence365/GroundVTS.

Video Large Language Models (VideoLLMs) have demonstrated impressive capabilities in video understanding, yet the massive number of input video tokens incurs a significant computational burden for deployment. Existing methods mainly prune video tokens at input level while neglecting the inherent information structure embedded in videos and large language models (LLMs). To address this, we propose HieraVid, a hierarchical pruning framework that progressively and dynamically reduces visual redundancy. Based on two observations that videos possess the segment-frame structure and LLMs internally propagate multi-modal information unidirectionally, we decompose pruning into three levels: 1) segment-level, where video tokens are first temporally segmented and spatially merged; 2) frame-level, where similar frames within the same segment are jointly pruned to preserve diversity; 3) layer-level, redundancy gradually shrinks as LLM layer increases w/o compromising performance. We conduct extensive experiments on four widely used video understanding benchmarks to comprehensively evaluate the effectiveness of HieraVid. Remarkably, with only 30% of tokens retained, HieraVid achieves new state-of-the-art performance, while maintaining over 98% and 99% of the performance of LLaVA-Video-7B and LLaVA-OneVision-7B, respectively.

Document understanding and GUI interaction are among the highest-value applications of Vision-Language Models (VLMs), yet they impose exceptionally heavy computational burden: fine-grained text and small UI elements demand high-resolution inputs that produce tens of thousands of visual tokens. We observe that this cost is largely wasteful -- across document and GUI benchmarks, only 22--71% of image patches are pixel-unique, the rest being exact duplicates of another patch in the same image. We propose PixelPrune, which exploits this pixel-level redundancy through predictive-coding-based compression, pruning redundant patches before the Vision Transformer (ViT) encoder. Because it operates in pixel space prior to any neural computation, PixelPrune accelerates both the ViT encoder and the downstream LLM, covering the full inference pipeline. The method is training-free, requires no learnable parameters, and supports pixel-lossless compression ($τ{=}0$) as well as controlled lossy compression ($τ{>}0$). Experiments across three model scales and document and GUI benchmarks show that PixelPrune maintains competitive task accuracy while delivering up to 4.2$\times$ inference speedup and 1.9$\times$ training acceleration. Code is available at https://github.com/OPPO-Mente-Lab/PixelPrune.

Vision Transformers (ViTs) have demonstrated state-ofthe-art performance in several benchmarks, yet their high computational costs hinders their practical deployment. Patch Pruning offers significant savings, but existing approaches restrict token reduction to deeper layers, leaving early-stage compression unexplored. This limits their potential for holistic efficiency. In this work, we present a novel Video Patch Pruning framework (VPP) that integrates temporal prior knowledge to enable efficient sparsity within early ViT layers. Our approach is motivated by the observation that prior features extracted from deeper layers exhibit strong foreground selectivity. Therefore we propose a fully differentiable module for temporal mapping to accurately select the most relevant patches in early network stages. Notably, the proposed method enables a patch reduction of up to 60% in dense prediction tasks, exceeding the capabilities of conventional image-based patch pruning, which typically operate around a 30% patch sparsity. VPP excels the high-sparsity regime, sustaining remarkable performance even when patch usage is reduced below 55%. Specifically, it preserves stable results with a maximal performance drop of 0.6% on the Youtube-VIS 2021 dataset.

Large Vision Language Models show impressive performance across image and video understanding tasks, yet their computational cost grows rapidly with the number of visual tokens. Existing token pruning methods mitigate this issue through empirical approaches while overlooking the internal mechanism of attention. In this paper, we propose a novel training free token pruning framework grounded in the dual form perspective of attention. We reformulate attention as an implicit linear layer whose weight matrix is the sum of rank 1 outer products, each generated by a single token's key value pair. Token pruning thus reduces to selecting an optimal subset of these rank 1 updates that best approximates the original dual weight matrix. Extending this perspective to standard softmax attention in LVLMs, we derive a novel metric quantifying both a token's information magnitude and information duplication. To efficiently select the subset with the proposed metric, we introduce Progressive Chunked Maximal Marginal Relevance. Extensive experiments demonstrate that our method achieves a better trade off between performance and efficiency, while providing another perspective on existing pruning approaches.

Vision-language models (VLMs) have been widely adopted for 3D question answering (3D QA). In typical pipelines, visual tokens extracted from multiple viewpoints are concatenated with language tokens and jointly processed by a large language model (LLM) for inference. However, aggregating multi-view observations inevitably introduces severe token redundancy, leading to an overly large visual token set that significantly hinders inference efficiency under constrained token budgets. Visual token pruning has emerged as a prevalent strategy to address this issue. Nevertheless, most existing pruners are primarily tailored to 2D inputs or rely on indirect geometric cues, which limits their ability to explicitly retain semantically critical objects and maintain sufficient spatial coverage for robust 3D reasoning. In this paper, we propose SeGPruner, a semantic-aware and geometry-guided token reduction framework for efficient 3D QA with multi-view images. Specifically, SeGPruner first preserves semantically salient tokens through an attention-based importance module (Saliency-aware Token Selector), ensuring that object-critical evidence is retained. It then complements these tokens with spatially diverse ones via a geometry-guided selector (Geometry-aware Token Diversifier), which jointly considers semantic relevance and 3D geometric distance. This cooperation between saliency preservation and geometry-guided diversification balances object-level evidence and global scene coverage under aggressive token reduction. Extensive experiments on ScanQA and OpenEQA demonstrate that SeGPruner substantially improves inference efficiency, reducing the visual token budget by 91% and inference latency by 86%, while maintaining competitive performance in 3D reasoning tasks.

Vision Transformers (ViTs) achieve state-of-the-art performance but suffer from the $O(N^2)$ complexity of self-attention, making inference costly for high-resolution inputs. To address this bottleneck, token pruning has emerged as a critical technique to accelerate inference. Most existing methods rely on the [CLS] token to estimate patch importance. However, we argue that the [CLS] token can be unreliable in early layers where semantic representations are still immature. As a result, pruning in the early layer often leads to inaccurate importance estimation and unnecessary information loss. In this work, we propose a training-free token importance metric, namely Col-Ln, which is derived from Rényi entropy that enables the identification of informative tokens from the first layer of the network, thereby enabling more reliable pruning in token reduction. Extensive experiments on ViTs and Large Vision-Language Models (LVLMs) demonstrate that our approach consistently outperforms state-of-the-art pruning methods across diverse benchmarks.

Edge detection is a fundamental image analysis task that underpins numerous high-level vision applications. Recent advances in Transformer architectures have significantly improved edge quality by capturing long-range dependencies, but this often comes with computational overhead. Achieving higher pixel-level accuracy requires increased input resolution, further escalating computational cost and limiting practical deployment. Building on the strong representational capacity of recent Transformer-based edge detectors, we propose an Adaptive Multi-stage non-edge Pruning framework for Edge Detection(Amped). Amped identifies high-confidence non-edge tokens and removes them as early as possible to substantially reduce computation, thus retaining high accuracy while cutting GFLOPs and accelerating inference with minimal performance loss. Moreover, to mitigate the structural complexity of existing edge detection networks and facilitate their integration into real-world systems, we introduce a simple yet high-performance Transformer-based model, termed Streamline Edge Detector(SED). Applied to both existing detectors and our SED, the proposed pruning strategy provides a favorable balance between accuracy and efficiency-reducing GFLOPs by up to 40% with only a 0.4% drop in ODS F-measure. In addition, despite its simplicity, SED achieves a state-of-the-art ODS F-measure of 86.5%. The code will be released.

Recent advances in large-scale video models have significantly improved video understanding across domains such as surveillance, healthcare, and entertainment. However, these models also amplify privacy risks by encoding sensitive attributes, including facial identity, race, and gender. While image anonymization has been extensively studied, video anonymization remains relatively underexplored, even though modern video models can leverage spatiotemporal motion patterns as biometric identifiers. To address this challenge, we propose a novel attention-driven spatiotemporal video anonymization framework based on systematic disentanglement of utility and privacy features. Our key insight is that attention mechanisms in Vision Transformers (ViTs) can be explicitly structured to separate action-relevant information from privacy-sensitive content. Building on this insight, we introduce two task-specific classification tokens, an action CLS token and a privacy CLS token, that learn complementary representations within a shared Transformer backbone. We contrast their attention distributions to compute a utility-privacy score for each spatiotemporal tubelet, and keep the top-k tubelets with the highest scores. This selectively prunes tubelets dominated by privacy cues while preserving those most critical for action recognition. Extensive experiments demonstrate that our approach maintains action recognition performance comparable to models trained on raw videos, while substantially reducing privacy leakage. These results indicate that attention-driven spatiotemporal pruning offers an effective and principled solution for privacy-preserving video analytics.

In recent years, GUI agents have demonstrated strong potential in navigation tasks. However, preserving complete historical screenshots introduces substantial computational overhead. This paper investigates how token pruning, a plug-and-play inference acceleration technique, can be effectively applied to GUI agent navigation scenarios. Firstly, we address the question of where pruning should occur. We identify a system-level redundancy overlooked by existing methods: as the same screenshot is repeatedly fed into the model across different steps, its ViT encoding is redundantly recomputed each time. We show that its ViT-encoded embeddings can be fully cached and reused across steps, substantially reducing FLOPs while preserving model performance. This finding suggests that inference acceleration efforts should focus on the subsequent Large Language Model (LLM). Building on this, we further address the question of how to prune within the LLM, and distill two key insights: (i) from a semantic perspective, the token budget should be balanced between foreground and background regions; (ii) from a spatial perspective, the spatial uniformity of retained tokens should be maintained to preserve the model's global spatial perception. These findings provide practical guidance for the design of inference acceleration and token pruning for GUI agent navigation.

The integration of Vision-Language-Action (VLA) models into autonomous driving systems offers a unified framework for interpreting complex scenes and executing control commands. However, the necessity to incorporate historical multi-view frames for accurate temporal reasoning imposes a severe computational burden, primarily driven by the quadratic complexity of self-attention mechanisms in Large Language Models (LLMs). To alleviate this bottleneck, we propose ETA-VLA, an Efficient Token Adaptation framework for VLA models. ETA-VLA processes the past $n$ frames of multi-view images and introduces a novel Intra-LLM Sparse Aggregator (ILSA). Drawing inspiration from human driver attention allocation, ILSA dynamically identifies and prunes redundant visual tokens guided by textual queries and temporal consistency. Specifically, we utilize a text-guided scoring mechanism alongside a diversity-preserving sparsification strategy to select a sparse subset of critical tokens, ensuring comprehensive awareness of the driving scene. Extensive experiments on the NAVSIM v2 demonstrate that ETA-VLA achieves driving performance comparable to state-of-the-art baselines while reducing computational FLOPs by approximately 32%. Notably, our method prunes 85% of visual tokens and reduces inference FLOPs by 61%, but still retaining 94% of the original accuracy on the NAVSIM v2 benchmark.

Vision-Language-Action (VLA) models excel in robotic manipulation but suffer from significant inference latency due to processing dense visual tokens. Existing token reduction methods predominantly rely on attention magnitude as a static selection. In this work, we challenge this assumption, revealing that high-attention tokens are task-dependent and can even degrade policy performance. To address this, we introduce TIES (Tau-guided Inter-layer Efficient Selection), a dynamic framework guided by inter-layer token ranking consistency. By adaptively balancing attention magnitude with ranking consistency, TIES ensures robust token selection without requiring additional training. On the CogACT + SIMPLER benchmark, TIES improves average success rates by 6% while reducing token usage by 78%, and demonstrate strong generalization across diverse decoders and benchmarks.

Recent multimodal large language models are computationally expensive because Transformers must process a large number of visual tokens. We present ReDiPrune, a training-free token pruning method applied before the vision-language projector, where visual features remain rich and discriminative. Unlike post-projection pruning methods that operate on compressed representations, ReDiPrune selects informative tokens directly from vision encoder outputs, preserving fine-grained spatial and semantic cues. Each token is scored by a lightweight rule that jointly consider text-conditioned relevance and max-min diversity, ensuring the selected tokens are both query-relevant and non-redundant. ReDiPrune is fully plug-and-play, requiring no retraining or architectural modifications, and can be seamlessly inserted between the encoder and projector. Across four video and five image benchmarks, it consistently improves the accuracy-efficiency trade-off. For example, on EgoSchema with LLaVA-NeXT-Video-7B, retaining only 15% of visual tokens yields a +2.0% absolute accuracy gain while reducing computation by more than $6\times$ in TFLOPs. Code is available at https://github.com/UA-CVML/ReDiPrune.

Document parsing is a fine-grained task where image resolution significantly impacts performance. While advanced research leveraging vision-language models benefits from high-resolution input to boost model performance, this often leads to a quadratic increase in the number of vision tokens and significantly raises computational costs. We attribute this inefficiency to substantial visual regions redundancy in document images, like background. To tackle this, we propose PaddleOCR-VL, a novel coarse-to-fine architecture that focuses on semantically relevant regions while suppressing redundant ones, thereby improving both efficiency and performance. Specifically, we introduce a lightweight Valid Region Focus Module (VRFM) which leverages localization and contextual relationship prediction capabilities to identify valid vision tokens. Subsequently, we design and train a compact yet powerful 0.9B vision-language model (PaddleOCR-VL-0.9B) to perform detailed recognition, guided by VRFM outputs to avoid direct processing of the entire large image. Extensive experiments demonstrate that PaddleOCR-VL achieves state-of-the-art performance in both page-level parsing and element-level recognition. It significantly outperforms existing solutions, exhibits strong competitiveness against top-tier VLMs, and delivers fast inference while utilizing substantially fewer vision tokens and parameters, highlighting the effectiveness of targeted coarse-to-fine parsing for accurate and efficient document understanding. The source code and models are publicly available at https://github.com/PaddlePaddle/PaddleOCR.

Efficient visual representation is a central image-processing challenge in embodied manipulation, where policies repeatedly process dense visual-token sequences during closed-loop control. Existing methods rank or prune tokens using semantic relevance, VLM attention, cross-frame redundancy, or motion in the action space. These signals may discard task-relevant regions when instruction-related appearance and observed image motion are not yet spatially aligned. We introduce Interaction-Aligned Pruning (IAprune), a training-free method that treats per-frame budget setting and within-budget token selection as two linked decisions. Semantic--motion spatial agreement guides the choice between Conservative and Aggressive coverage, and the resulting region size is mapped to a calibrated dynamic budget. Continuous semantic and motion responses rank the existing tokens, while geometric residual correction redirects fixed selection slots toward under-represented boundaries without increasing the sequence length. Across four embodied manipulation policies, three simulation benchmarks, and a real-robot platform, IAprune provides a favorable accuracy--efficiency trade-off, matching the unpruned policy on LIBERO with a $1.54\times$ speedup and reaching $1.48\times$ acceleration on a real robot. Phase-wise analysis shows that retention gains are largest under tight budgets early in an episode, while fixed-budget analysis confirms that geometry replaces low-priority tokens with boundary and contact-region evidence rather than retaining more tokens. Our project website is: IAprune.com.

Due to the great saving of computation and memory overhead, token compression has become a research hot-spot for MLLMs and achieved remarkable progress in image-language tasks. However, for the video, existing methods still fall short of high-ratio token compression. We attribute this shortcoming to the insufficient modeling of temporal and continual video content, and propose a novel and training-free token pruning method for video MLLMs, termed ForestPrune, which achieves effective and high-ratio pruning via Spatial-temporal Forest Modeling. In practice, ForestPrune construct token forests across video frames based on the semantic, spatial and temporal constraints, making an overall comprehension of videos. Afterwards, ForestPrune evaluates the importance of token trees and nodes based on tree depth and node roles, thereby obtaining a globally optimal pruning decision. To validate ForestPrune, we apply it to two representative video MLLMs, namely LLaVA-Video and LLaVA-OneVision, and conduct extensive experiments on a bunch of video benchmarks. The experimental results not only show the great effectiveness for video MLLMs, e.g., retaining 95.8% average accuracy while reducing 90% tokens for LLaVA-OneVision, but also show its superior performance and efficiency than the compared token compression methods, e.g., +10.1% accuracy on MLVU and -81.4% pruning time than FrameFusion on LLaVA-Video.

Large Vision-Language Models (LVLMs) excel in visual understanding and reasoning, but the excessive visual tokens lead to high inference costs. Although recent token reduction methods mitigate this issue, they mainly target single-turn Visual Question Answering (VQA), leaving the more practical multi-turn VQA (MT-VQA) scenario largely unexplored. MT-VQA introduces additional challenges, as subsequent questions are unknown beforehand and may refer to arbitrary image regions, making existing reduction strategies ineffective. Specifically, current approaches fall into two categories: prompt-dependent methods, which bias toward the initial text prompt and discard information useful for subsequent turns; prompt-agnostic ones, which, though technically applicable to multi-turn settings, rely on heuristic reduction metrics such as attention scores, leading to suboptimal performance. In this paper, we propose a learning-based prompt-agnostic method, termed MetaCompress, overcoming the limitations of heuristic designs. We begin by formulating token reduction as a learnable compression mapping, unifying existing formats such as pruning and merging into a single learning objective. Upon this formulation, we introduce a data-efficient training paradigm capable of learning optimal compression mappings with limited computational costs. Extensive experiments on MT-VQA benchmarks and across multiple LVLM architectures demonstrate that MetaCompress achieves superior efficiency-accuracy trade-offs while maintaining strong generalization across dialogue turns. Our code is available at https://github.com/MArSha1147/MetaCompress.

Multimodal large language models suffer from severe computational and memory bottlenecks, as the number of visual tokens far exceeds that of textual tokens. While recent methods employ projector modules to align and compress visual tokens into text-aligned features, they typically depend on fixed heuristics that limit adaptability across diverse scenarios. In this paper, we first propose Query Guided Mixture-of-Projector (QMoP), a novel and flexible framework that adaptively compresses visual tokens via three collaborative branches: (1) a pooling-based branch for coarse-grained global semantics, (2) a resampler branch for extracting high-level semantic representations, and (3) a pruning-based branch for fine-grained token selection to preserve critical visual detail. To adaptively coordinate these branches, we introduce the Query Guided Router (QGR), which dynamically selects and weights the outputs from different branches based on both visual input and textual queries. A Mixture-of-Experts-style fusion mechanism is designed to aggregate the outputs, harnessing the strengths of each strategy while suppressing noise. To systematically evaluate the effects of Visual Token Compression, we also develop VTCBench, a dedicated benchmark for evaluating the information loss induced by visual token compression. Extensive experiments demonstrate that despite relying on fundamental compression modules, QMoP outperforms strong baselines and delivers significant savings in memory, computation, and inference time.

Large Vision-Language Models (LVLMs) rely on dense visual tokens to capture fine-grained visual information, but processing all these tokens incurs substantial computational and memory overhead during inference. To address this issue, we propose ResPrune, a training-free visual token pruning framework that enables efficient LVLM inference by selecting a compact yet informative subset of visual tokens. ResPrune formulates visual token pruning as a subspace reconstruction problem and employs a greedy subspace expansion strategy guided by residual energy, allowing it to preserve the geometric structure of the original visual token space. To further incorporate cross modal alignment, the selection process is conditioned on textual relevance, encouraging the retention of tokens that are both informative and instruction-relevant. The proposed method is lightweight and model-agnostic, and can be seamlessly integrated into existing LVLM pipelines without retraining or architectural modifications. Extensive experiments on multiple LVLM backbones, including LLaVA-1.5, LLaVA-NeXT, and Qwen2.5-VL, demonstrate that ResPrune consistently outperforms existing pruning approaches across a wide range of benchmarks, while achieving effective reductions in computation, memory consumption, and inference latency.

Although current Video-LLMs achieve impressive performance in video understanding tasks, their autoregressive decoding efficiency remains constrained by the massive number of video tokens. Visual token pruning can partially ease this bottleneck, yet existing approaches still suffer from information loss and yield only modest acceleration in decoding. In this paper, we propose ParallelVLM, a training-free draft-then-verify speculative decoding framework that overcomes both mutual waiting and limited speedup-ratio problems between draft and target models in long-video settings. ParallelVLM features two parallelized stages that maximize hardware utilization and incorporate an Unbiased Verifier-Guided Pruning strategy to better align the draft and target models by eliminating the positional bias in attention-guided pruning. Extensive experiments demonstrate that ParallelVLM effectively expands the draft window by $1.6\sim1.8\times$ with high accepted lengths, and accelerates various video understanding benchmarks by 3.36$\times$ on LLaVA-Onevision-72B and 2.42$\times$ on Qwen2.5-VL-32B compared with vanilla autoregressive decoding.

Token pruning is essential for enhancing the computational efficiency of vision-language models (VLMs), particularly for video-based tasks where temporal redundancy is prevalent. Prior approaches typically prune tokens either (1) within the vision transformer (ViT) exclusively for unimodal perception tasks such as action recognition and object segmentation, without adapting to downstream vision-language tasks; or (2) only within the LLM while leaving the ViT output intact, often requiring complex text-conditioned token selection mechanisms. In this paper, we introduce Spatio-Temporal Token Scoring (STTS), a simple and lightweight module that prunes vision tokens across both the ViT and the LLM without text conditioning or token merging, and is fully compatible with end-to-end training. By learning how to score temporally via an auxiliary loss and spatially via LLM downstream gradients, aided by our efficient packing algorithm, STTS prunes 50% of vision tokens throughout the entire architecture, resulting in a 62% improvement in efficiency during both training and inference with only a 0.7% drop in average performance across 13 short and long video QA tasks. Efficiency gains increase with more sampled frames per video. Applying test-time scaling for long-video QA further yields performance gains of 0.5-1% compared to the baseline. Overall, STTS represents a novel, simple yet effective technique for unified, architecture-wide vision token pruning.