Overview

Visual token pruning studies how to remove or merge image and video patches before or inside vision transformers and vision-language models, reducing computation and memory for tasks such as visual question answering, video understanding, document reading, and robot control. Methods use attention or text-query relevance, feature diversity and spatial coverage, and signals such as entropy, spectral structure, or geometry; learned selectors and adaptive budgets choose different tokens or compression rates for different inputs. Pruning can happen before visual encoding, across language-model layers, or during video processing, and is evaluated on both task benchmarks and practical measures such as prefill latency, FLOPs, and memory. The central tension is that a token that seems redundant early may become important for later reasoning, while selection overhead and hardware execution can prevent fewer tokens from yielding faster end-to-end inference.

Latest papers

298 papers23 papers in the last four weeks, down 48% on the four weeks before. 0.2% of all new papers.

Nov 17, 2026

LoopZip: Visual Token Compression for Efficient Looped Vision-Language Models

Looped vision-language models (VLMs) deepen computation by reusing shared layers, approaching larger models' capabilities with fewer parameters and easing the storage and deployment burden of stacking independent layers. However, repeatedly processing long visual sequences still incurs substantial compute and GPU memory costs. Visual token compression reduces these costs by shortening sequences. Existing methods mainly select original tokens directly or combine selection with aggregation. In looped models, the former may overlook changing token importance across calls and discard information needed later; the latter lack joint consideration of output changes across calls, limiting information preservation. We propose LoopZip, a training-free visual token compression algorithm addressing both issues. Visual aggregation builds a compact candidate pool before the looped backbone, and cross-invocation selection averages instruction attention from the first two existing calls of the same shared attention layer to retain tokens needed by both states. Weight-tied value correction then jointly fits shared visual-value adjustments to both pre-selection layer outputs, reducing output deviations across states. To validate LoopZip's effectiveness, we conduct extensive experiments covering two open-source looped VLMs, eleven benchmarks, and three pruning ratios. On LoopVL-1B, LoopZip retains 97.16% of the original model's performance at 50% pruning ratio and exceeds the strongest baseline by 5.36% at 87.5% pruning ratio. Across Hugging Face Transformers, vLLM, and SGLang, LoopVL-1B with LoopZip reduces peak allocated GPU memory by 58.36%-58.98% and achieves 2.01-2.47x end-to-end speedups over the original model.
Nov 17, 2026

PCRG: Training-Free Pruning and Caching of Reference Tokens for Video Generation

Reference-to-video diffusion transformers pack the variational autoencoder (VAE) latents of all references and the generated tokens into one full-attention sequence, in which reference tokens make up about half. Because attention cost grows quadratically with sequence length, these tokens take up an even larger share of the attention computation, yet most of them are static, smooth or irrelevant to the output. To accelerate reference-conditioned generation, we propose pruning and caching of reference tokens for video generation (PCRG), a training-free, plug-and-play method that uses two signals the model already computes to assign each reference token to one of three tiers, namely pruned, cached or kept. The first is a low-level redundancy r from VAE latents, since a token whose latent barely changes over time or across its neighbors adds little new content. The second is a high-level importance g from the vision-language-model (VLM) conditioner, where the attention that the prompt and any first or last frames pay to a token tells whether the output needs to read it. Tokens that are not redundant are kept, and redundant tokens are cached if important and pruned otherwise. Pruned tokens leave the sequence before the first step with all rotary position embedding positions unchanged. Cached tokens have their keys and values computed once and reused, and kept tokens are processed as usual. On MiniMax-H3, already distilled to 8 timesteps and quantized to int8, PCRG is 1.57x faster end to end on our V2V-Bench and 1.49x faster on the public VACE-Benchmark, while the quality of VBench drops by only 0.24% and 0.17%, respectively. PCRG shortens the sequence rather than changing how attention is computed or which steps are run, so it also stacks on step caching and sparse attention. On V2V-Bench it reaches 1.86x with TeaCache, 1.82x with Spark at nearly unchanged fidelity and 2.15x with Veda.
Nov 17, 2026

Disagree to Prune: Reallocating Visual Tokens Where a Small VLM Dissents

Visual token pruning speeds up large vision-language models (VLMs) by keeping only a part of their visual tokens. Most existing methods improve the ranking of the tokens and then answer every question once with the same budget. With strong rankings, pruning hurts only a few questions, and signals computed on the kept tokens see little of what it removed. We find that, on most of these questions, a small VLM given the full image disagrees with the pruned answer, since it still sees the missing evidence. We propose PRUDENT, which first answers with about half the budget and spends the saved tokens on re-answering the disagreeing questions at twice the budget. We show when such re-answering pays and why the small VLM should read the full image. On 18 benchmarks, PRUDENT outperforms eight existing pruning methods at the same average budget of visual tokens. Its lead widens as the budget tightens, and holds across five InternVL sizes. PRUDENT also has the highest average in two more model families. The disagreement of the small VLM flags most pruning failures of all eight methods, and re-answering on it also improves the two strongest with their own rankings. Other checkers give similar gains, and the extra tokens go where the first pass hurts the answer. These results suggest that identifying which pruned answers are trustworthy provides an additional lever for visual token pruning beyond the ranking of tokens. Code and data are available at https://github.com/LiZaiyuan0619/PRUDENT.
Nov 16, 2026

POKE: Probe Offline, Keep Efficiently for Vision-Language-Action Models

Visual token pruning reduces the inference cost of Vision-Language-Action (VLA) models, but attention-based scores do not directly measure a token's effect on the predicted action. We define action sensitivity as the magnitude of the first-order action change when a token is replaced by its camera's mean token and estimate it for all tokens with a single backward pass. Computing it online is costly: scoring on the full (unpruned) graph needs an extra forward pass at every control step. Scoring on the pruned forward pass is cheaper but causes scoring-graph lock-in: removed tokens lose their direct paths through later layers and tend to receive lower scores, making those locations less likely to return. On replayed observations, even with a random conditioning mask, pruned-graph scoring re-selects 59% of the mask's tokens, versus 6% expected by chance; attention histories show a similar bias. POKE (Probe Offline, Keep Efficiently) computes full-graph action sensitivities offline and distills them into a student with under 0.3M parameters. At inference, the student scores the current observation without backpropagation or previous masks, and score-guided diversity selection retains the tokens. Keeping 32 of 512 visual tokens on LIBERO, POKE achieves 95.2% success on OpenVLA-OFT and 90.8% on π0.5π_{0.5} at 1.85×1.85\times and 1.26×1.26\times speedups, outperforming attention-, diversity-, and distillation-based pruners, with larger margins on LIBERO-Plus.
Nov 16, 2026

How Reliable Is Visual Token Selection? Rethinking Pruning in Multimodal LLMs

Visual token pruning improves the efficiency of multimodal large language models (MLLMs) by removing redundant visual tokens, yet the reliability of existing selection signals has not been systematically examined. We analyze token selection under semantics-preserving perturbations and across representative pruning criteria, revealing two limitations. First, attention-based importance shifts non-uniformly under perturbations, so magnitude alone cannot distinguish stable evidence from unstable saliency (local instability). Second, attention- and diversity-based methods select substantially different subsets despite comparable performance, showing that useful evidence is not confined to a unique subset (global inconsistency). These findings motivate reliable token-set construction. We instantiate this perspective in CoRePruner, a training-free framework combining instruction-aware diversity with reliability-aware attention. Across image benchmarks, multiple MLLM architectures, and token budgets, CoRePruner improves performance, especially under aggressive pruning, and remains competitive on video benchmarks. Our results establish reliability as a central consideration in visual token selection.
Nov 15, 2026

Glance, then Select: Evidence-Tempered Visual Token Selection for Extremely Compressed Multimodal LLMs

Multimodal large language models (MLLMs) encode a high-resolution image into thousands of visual tokens, which dominate the cost of the prefill stage. Visual token reduction removes most of these tokens, but two problems remain at extreme budgets. The vision encoder still processes every patch of the image, and the few surviving tokens often miss the small regions that hold the answer. In this paper, we propose Glance-ETD, a two-stage token reduction framework guided by a lightweight question-conditioned scorer. In the first stage, a glance encoder views the image at low resolution and re-encodes only the question-relevant regions at native resolution. In the second stage, Evidence-Tempered Diversity (ETD) selects the final tokens by combining diversity with the evidence predicted by the scorer, so that the kept tokens cover the scene and retain the answer. A single-rollout self-distillation recipe trains the MLLM with the glance encoder. With only 5% of the visual tokens, Glance-ETD retains 91.6% of the full-token performance of Qwen3.5-4B across fourteen benchmarks and outperforms the strongest prior method, LT-OPD, by 17.3 points and the same training recipe with DivPrune by 6.0. It also speeds up the time to first token on 4K images by 3.1x and generalizes to three other MLLMs and to high-resolution document benchmarks, with the largest gains on text-rich tasks.
Nov 15, 2026

TIVS: Mapping the Sensitivity Landscape of Visual Token Transformations in MLLMs

Recent efforts to improve the efficiency of multimodal large language models (MLLMs) have explored visual token pruning, quantization, and feature compression. However, the properties that determine whether transformed visual tokens remain usable by a frozen decoder are not well characterized. We introduce TIVS (Token Identity Versus Statistics), a training-free intervention framework that separately studies sequence permutation, token retention, numerical precision, and statistical replacement under controlled conditions. On 192 POPE questions, three token permutations change TinyLLaVA accuracy by only −0.17-0.17 percentage points on average, while changing 4.69% of individual predictions. Diagonal and rank-64 Gaussian replacements decrease accuracy by 25.0 and 28.1 points on the same subset. Further experiments with exact moment constraints, quantization error, token retention, and additional visual tasks reveal how strongly the effects depend on the original feature values and the evaluation setting. These findings provide an empirical basis for assessing visual token transformations in frozen MLLMs and designing more reliable multimodal interfaces.
Nov 12, 2026

DRAMA: Dissecting Attention Redundancy for Accelerating Multimodal Diffusion Large Language Models

Multimodal diffusion large language models (dLLMs) generate text through iterative masked denoising with bidirectional attention, offering a compelling alternative to autoregressive vision-language modeling. However, each denoising step requires a full forward pass over long multimodal sequences in which visual tokens dominate the sequence length, and a fixed step budget further compounds this quadratic per-step cost. Existing inference acceleration approaches for multimodal dLLMs characterize attention redundancy implicitly, motivating designs from isolated attention visualizations rather than from measurement. We close this gap with DRAMA, a training-free acceleration framework grounded in a systematic dissection of attention redundancy in multimodal dLLMs. By probing attention tensors along spatial and temporal axes, we reveal two persistent redundancy structures: (i) attention mass consistently concentrates on a sparse subset of visual tokens, and (ii) attention distributions rapidly stabilize across denoising steps after an initial formation phase. These findings motivate two complementary mechanisms: Spatial-Redundancy-Aware Visual Token Pruning (SVTP) and Sufficiency-Guided Adaptive Stopping (SGAS). Together, DRAMA achieves an average 18.67 times inference speedup across six vision-language benchmarks without training, architectural modification, or attention caching, while maintaining competitive accuracy. Our results establish a strong, measurement-grounded acceleration baseline for multimodal dLLM inference.
Nov 11, 2026

DiFiT: Distribution-Fidelity-guided Visual Token Pruning for Multimodal Large Language Models

Visual token pruning reduces the inference cost of multimodal large language models, but answer accuracy only partly reflects its effect. Pruning can change the probabilities of alternative response tokens even when greedy answers remain correct and unchanged. We compare full and pruned response distributions under shared reference prefixes. Average predictive drift grows with compression on fixed correctly answered samples. Separate analyses associate larger drift with more confidence-decision disagreements and sampling failures. We propose DiFiT (Distribution-Fidelity-guided Token pruning) to learn discrete selection from reference answers and full-token response distributions. Dual-View Distillation matches probabilities across the vocabulary, while answer supervision supports task quality. A learned scorer and differentiable Top-KK learn which visual evidence to retain. A multi-budget mask objective supports different retention ratios through one shared ranking over a frozen model. On Qwen2.5-VL-7B, DiFiT reduces sampling-induced negative flips by 30.3% relative to FastV at 10% retention on four-task greedy-correct controls. It preserves 98.74% of baseline performance on MME at 30% retention.
Nov 9, 2026

APS-VLA: Action-Guided and Future-Aware Token Pruning for Efficient Vision-Language-Action Models

Vision-language-action (VLA) policies have demonstrated strong generalization in robotic manipulation, yet processing dense visual tokens often incurs high inference latency. Visual token pruning offers a practical way to reduce this cost, but identifying a compact set of tokens that preserves reliable action generation remains challenging. Existing methods largely rely on static or immediate heuristic cues to estimate token importance. We introduce APS-VLA, a framework that learns visual token selection from robot demonstrations while keeping the pretrained VLA backbone frozen. A lightweight Decision Influence Estimator scores tokens conditioned on current visual observations, language instructions, and robot proprioception. Its training combines two complementary signals: action supervision matches both the full-input policy predictions and demonstrated actions, while an auxiliary visual transition objective uses demonstrated actions to predict subsequent visual features from the selected representation. At deployment, the estimator performs a single hard selection and packing step before the policy's native action-generation procedure; the auxiliary predictor is removed. Experiments with OpenVLA-OFT on LIBERO and with π0.5π_{0.5} and XR-1 on RoboCasa365 demonstrate inference acceleration with modest changes in task success. Across five real-world tasks on two dual-arm platforms, APS-VLA reduces mean policy inference latency from 137.25 to 93.25 ms, achieving a 1.47×1.47\times speedup with an average success rate of 61%, compared with 60% for the unpruned baseline. Together, these results support APS-VLA as a modular, plug-in framework for efficient robotic control across VLA architectures, preserving the underlying policy parameters and native action-generation procedure.
Nov 6, 2026

ActPrune: Action-Grounded Visual Token Pruning for Efficient Vision-Language-Action Models

Vision-language-action (VLA) models perform versatile manipulation tasks from language instructions, but their inference cost limits how fast a robot can respond in closed-loop control. Visual token pruning reduces this cost; however, prior methods estimate token redundancy from encoded visual embeddings or attention, only after the visual encoder has processed the entire observation. We propose ActPrune, a training-free token pruning method that moves the pruning decision ahead of the vision encoder by grounding it in a geometric prior derived from the robot's own action. The action chunk predicted at the previous control step is already available before the current observation is encoded, and it specifies the end-effector and camera motion between consecutive observations. The same action affects the two standard camera views asymmetrically: in the fixed third-person view, ActPrune retains patches along the projected end-effector trajectory together with semantic anchors, while in the wrist view, it prunes patches that remain unchanged after compensating for the action-induced camera motion. Extensive experiments on two simulation environments (LIBERO and VLABench) and four real-robot tasks demonstrate that ActPrune achieves up to 1.61×\times speedup and 59.5% FLOPs reduction with negligible degradation in success rates. The project site is http://mmai.ewha.ac.kr/actprune/.
Nov 5, 2026

T3R: Training-Free Two-Stage Token Refinement Towards Efficient and Robust VLA Models

Vision-Language-Action (VLA) models enable end-to-end robot control, but inference is increasingly expensive. Each decision processes hundreds of visual tokens, often across multiple cameras, leading to substantial computation while allowing irrelevant background content to distract action prediction under distribution shift. We argue that these issues stem from two complementary forms of redundancy: many visual tokens are spatially irrelevant, while attention over the remaining tokens is often misallocated. Based on this observation, we propose T3R, a Training-Free Two-Stage Token Refinement framework. In the first stage, segmentation-guided token pruning removes irrelevant visual patches to reduce computation and eliminate distracting context. In the second stage, saliency-guided attention biasing reallocates attention toward action-critical visual regions and language instructions within the retained tokens. Together, pruning produces a cleaner candidate set, while attention biasing further emphasizes the most decision-relevant evidence, yielding a better accuracy-efficiency frontier than either stage alone. On LIBERO-Spatial, T3R removes 85% of scene-camera visual tokens while essentially matching the unpruned success rate, and improves success by up to 7 percentage points under out-of-distribution object-placement perturbations. The timing comparison attributes 7.8 ms per inference call to pruning and attention-bias application. As input resolution increases, LLM-forward profiling shows up to a 3.3x speedup over the unpruned baseline. Evaluations across LIBERO, SIMPLER, and RoboTwin with three VLA backbones show the strongest accuracy-efficiency benefits in multi-camera settings. These results demonstrate that training-free token refinement is a practical approach to efficient and robust VLA inference.
Nov 5, 2026

Training-Free Predictive Spatio-Temporal Token Pruning for Few-Step Autoregressive Video Diffusion

Few-step autoregressive (AR) video diffusion models generate videos chunk by chunk and have made efficient video generation practical. However, further accelerating these models toward real-time generation remains challenging, as few-step distillation leaves little cross-step redundancy for existing training-free caching methods to exploit. We revisit this regime and identify two token-level redundancy patterns: cross-frame redundancy from static content shared across adjacent frames, and cross-step redundancy from tokens that converge early during denoising. Crucially, we find that early clean estimates along the model's own denoising trajectory provide an effective online signal for locating both patterns, requiring no retraining, auxiliary predictor, or reference video. Based on these findings, we propose PreSTP (Predictive Spatio-Temporal Token Pruning), a training-free framework that constructs a unified token-wise redundancy mask from cross-frame similarity and cross-step convergence. Tokens marked as redundant skip subsequent computation under a consistent last-computed-value restoration rule, and a content-adaptive mask reuse strategy further reduces dense probing cost. On 5-second generation, PreSTP achieves 1.52x-3.20x speedups and a better efficiency-fidelity trade-off than coarse-grained caching baselines such as TeaCache and FlowCache. On 30-second generation, it reaches up to 1.96x speedup while improving VBench-Long and suppressing long-horizon drift. Notably, it continues to reduce inference cost even in the extreme one-step regime, where cross-step redundancy vanishes entirely and coarse step-wise caching is no longer applicable.
Nov 4, 2026

Visual Token Routing: Learning to keep tokens that matter

Computer-use agents (CUAs) have recently achieved superhuman performance on several long-horizon benchmarks. However, their computational cost limits their use in long-horizon tasks and real-time domains such as video games. A key bottleneck is representing each screenshot as thousands of visual tokens, which accumulate into long contexts as agents reason over extended interaction histories. We observe that most of these visual tokens are either semantically irrelevant to the agent's next action or temporally redundant across screenshots. Building on this observation, we introduce Visual Token Routing (VTR), a joint training framework for learning sparse visual representations that preserve the information needed for action prediction. In VTR, we train the vision-language model together with a lightweight spatio-temporal router that examines each image region in the current and previous screenshots and decides whether its visual token enters the language model. We optimize the router using an adaptive loss to keep the number of visual tokens as small as possible while minimizing performance degradation. With just 2,000 training steps, VTR matches the performance of the full-resolution model on OSWorld, CUA-World, and video game action prediction, while using 2.5-5.8x fewer visual tokens. These savings reduce inference cost by up to 3.3x and speed up prefill by up to 2.9x. VTR also outperforms lowering the image resolution and other baselines for dropping tokens, by up to 12.8 points on OSWorld. Overall, VTR enables computer-use agents to process only the visual tokens they need, allowing them to be used in long-horizon tasks and real-time domains at a fraction of the computational cost. Code and models are available at https://pranjal2041.github.io/vtr
Nov 4, 2026

RI-Prune: Relation and Information Aware Visual Token Pruning for Multi-Image Vision-Language Models

Large vision-language models (LVLMs) have recently been extended to handle multiple input images, but this extension dramatically increases the visual token budget, making visual-token pruning not merely an efficiency option but a practical necessity. Existing visual-token pruning methods, primarily designed for the single-image setting, rely on attention-derived saliency signals to estimate per-token importance. However, this attention-centric design transfers poorly to the multi-image regime: it causes substantial performance degradation on multi-image benchmarks, suffers from strong positional bias, and loses discriminative attention signals as the number of input images increases. To test whether attention-selected tokens are truly important for multi-image reasoning, we conduct two complementary causal probes Leave-One-In and Leave-One-Deletion, which measure token importance by restoring individual tokens to a zeroed visual sequence or removing them from the full sequence. Attention-based pruners provide only marginal gains over random selection, showing that attention is a weak proxy for visual-token importance in multi-image reasoning. Motivated by this gap, we propose Relation and Information Aware Pruning (RI-Prune), a training-free, attention-free pre-decoder selector for multi-image visual token pruning. RI-Prune scores each visual token using a z-normalized combination of three signals computed from projector outputs: query relevance, cross-image correspondence with compact per-image prototypes, and intra image representativeness. By integrating these signals, RI-Prune performs balanced per-image Top-K selection to preserve information across input images. Across multiple multi-image benchmarks and three LVLM backbones, RI-Prune achieves a stronger trade-off between token reduction and task performance than prior pruning methods, particularly under aggressive token budgets.
Nov 3, 2026

BtsPruner: Role-Aware Visual Token Pruning for Large Vision-Language Models

Large Vision-Language Models (LVLMs) often process hundreds or thousands of visual tokens, resulting in substantial computational and memory overhead during inference. Existing training-free pruning methods typically estimate token importance using language-model attention, visual attention, or feature similarity. However, these methods treat visual tokens with uniform criteria and overlook that tokens may serve different roles during visual encoding. In this work, we investigate the evolution of visual token roles in LVLMs and observe that tokens progressively transition from local detail carriers to globally referenced representatives as encoding proceeds. Based on this observation, we propose BtsPruner, a training-free role-aware visual token pruning framework that preserves complementary token roles under a fixed budget. Specifically, BtsPruner first identifies the token-role transition point based on the evolution of attention concentration. It then discovers globally referenced representatives from later encoding stages and recursively retrieves local detail tokens from earlier stages to recover complementary fine-grained evidence. Extensive experiments across LLaVA-1.5, LLaVA-NeXT, Qwen-VL, InternVL, and CogVLM demonstrate that BtsPruner achieves superior performance retention under aggressive token reduction. On LLaVA-1.5-7B, BtsPruner preserves 91.6% of the dense model's performance using only 32 of 576 visual tokens (5.6%). In the high-resolution LLaVA-NeXT-7B setting, it achieves up to a 5.95x CUDA-latency speedup.
Nov 1, 2026

SVD-Prune: Spectral Vision Token Pruning under Extreme Compression

Vision-Language Models (VLMs) and Vision-Language-Action (VLA) models process hundreds of visual tokens, making visual representations a major computational bottleneck. We argue that visual token pruning should be formulated not as identifying individually important tokens, but as preserving the dominant low-rank spectral structure of the visual representation. Based on this formulation, we introduce SVD-Prune, a training-free framework that leverages Singular Value Decomposition (SVD) and statistical leverage scores for globally informed token selection. SVD-Prune is architecture-agnostic and can be applied at different stages of the multimodal pipeline, including encoder-side, decoder-side, and hybrid configurations Extensive experiments across VLM and VLA benchmarks demonstrate strong robustness under extreme compression. SVD-Prune reduces the visual token. SVD-Prune cuts the visual token budget by up to 99.3%, while, at a 16-token budget, encoder-side pruning lowers inference FLOPs by 82.4% and achieves a 4.18x end-to-end speedup, including pruning overhead.
Oct 30, 2026

SIEVE: Preserving Cross-Image Evidence via Depth-Aware Token Compression

Multi-image vision-language models (VLMs) incur substantial computation when processing long visual sequences. Compressing these sequences must preserve complementary evidence across images and keep tokens from different images separate. Global selection can underrepresent question-relevant images, and pruning in early decoder layers risks removing relevant content before sufficient multimodal processing, yet existing methods spend most of the budget on these early layers. We propose SIEVE, a training-free, single-pass framework that jointly allocates visual tokens across images and decoder depth. SIEVE first pools each image into a compact spatial grid to reduce early processing cost. It then progressively selects representatives using prompt attention under equal per-image quotas and aggregates unselected tokens into similar representatives from the same image. These operations keep every image represented and its tokens separate from other images, while refining the retained representations. A closed-form schedule sets retention counts to target a prescribed average token budget across decoder layers. Across four VLM backbones, SIEVE achieves the highest average at a 6.25% budget on five multi-image benchmarks and four single-image benchmarks, and the highest mean on two video benchmarks. On LLaVA-OneVision-7B, it outperforms the strongest baseline by 9.96 and 9.03 points at 6.25% and 12.5% budgets, and at 12.5% it exceeds the uncompressed model by 2.01 points with 6.6x fewer prefill FLOPs.
Oct 29, 2026

Let LiDAR Decide Where to Look: Geometry-Measured Token Selection for Efficient Sparse Latent World Models

Vision-centric autonomous driving models encode thousands of image tokens per frame, yet most depict sky, buildings, or empty road---content irrelevant to planning. Worse, these tokens dilute the planner's attention: we show that randomly discarding 75% of tokens already outperforms the dense model, because the planner can concentrate on the remaining driving-critical content. Building on this finding, we present Geometry-Aware Token Selection (GATS), which uses LiDAR and radar point projections to select a small fraction of image tokens before the vision backbone runs. Because the selector operates on measured 3D geometry rather than visual features, the backbone encodes only the selected patches, cutting end-to-end FLOPs by up to 18×18\times. Each kept token receives a 3D anchor from its LiDAR point and is fed to a self-supervised latent world model that predicts future features and 3D displacements without box annotations. At 25% budget, GATS achieves a 3-sec L2 displacement error of 0.520 m on nuScenes and an EPDMS of 88.9 on NAVSIM---surpassing the dense baseline (0.573 m / 85.9) at 6.6×6.6\times fewer total FLOPs. At an extreme 10% budget (24 GFLOPs, 18×18\times reduction), it still beats the dense baseline on both nuScenes (0.537 m vs. 0.573 m L2) and NAVSIM (87.9 vs. 85.9 EPDMS). On 739 intent-critical scenarios, GATS covers 95.8% of hazardous vehicles and achieves a 0.09% collision rate---73.5% lower than the dense model (0.34%).
Oct 28, 2026

Don't Ask the LLM: Visual Token Pruning with Spectral Structure and Semantic Relevance

Large vision-language models (LVLMs) represent images and videos as long visual-token sequences, substantially increasing inference costs. These costs motivate visual-token pruning, whose effectiveness depends on both what signal is used for selection and when that signal becomes available. Our analysis reveals complementary limitations: visual-side [CLS] attention produces scattered, background-biased selections under aggressive compression, while stronger query-conditioned signals become available after substantial computation within the large language model (LLM). This raises a natural question: can structurally coherent, query-aware token selection be performed before consulting the LLM? We therefore propose ST-Pruner (Spectral-Textual Pruner), a training-free framework for pre-LLM visual-token selection. It combines attention-independent spectral saliency with inverted patch-text similarity through parameter-free multiplicative fusion. Across diverse LVLMs, ST-Pruner retains over 95% of full-token performance using at most 11.1% of visual tokens and achieves a 3.2x end-to-end speedup on LLaVA-NeXT. Code is available at https://github.com/sejong-rcv/ST-Pruner.
Oct 28, 2026

ConvRoute-ViT: Convolutional Token Routing for Efficient Vision Transformers

Vision Transformers (ViTs) achieve strong image recognition performance, but their self-attention increases quadratically with the number of input tokens and makes them computationally expensive. The standard implementation of ViT models treats all patch tokens with equal importance across every layer of the model. This uniform treatment introduces redundant computation and may reduce the model's focus on the most discriminative object regions. In this paper, we propose ConvRoute-ViT, a convolutional token routing mechanism for efficient vision transformers that performs token reduction through convolutional routing. Our model preserves the most relevant patch tokens according to class-token attention at selected layers, while the remaining tokens are spatially reorganized and compressed into a compact set of convolutional summary tokens. This design reduces the sequence length without completely removing the contextual information of the less informative tokens. By applying the routing process hierarchically, the proposed framework constructs compact and discriminative representations for deeper layers. Experiments on CIFAR-100 and TinyImageNet show that ConvRoute-ViT improves computational efficiency while maintaining classification accuracy compared with the base ViT.
Oct 27, 2026

LearnPrunerV2: Learnable Progressive Token Pruning for Vision-Language Models

Vision-Language Models (VLMs) have demonstrated remarkable multimodal capabilities but suffer from prohibitive computational overhead due to the massive number of visual tokens. While token pruning is a promising solution, existing methods still suffer from two major drawbacks: 1) unreliable pruning criteria that fail to accurately localize task-specific regions due to inherent attention biases or query-agnostic designs, and 2) suboptimal pruning strategies that employ rigid, non-adaptive schedules, overlooking the dynamic evolution of token redundancy across different model depths. To address these challenges, we propose LearnPrunerV2, which extends the previous two-stage heuristic framework (LearnPruner) into a fully learnable architecture. Specifically, we integrate lightweight learnable modules into every layer of the VLM backbone, which estimate token importance by fusing intrinsic visual saliency with extrinsic query relevance captured via a shadow token. By uncovering a consistent staircase redundancy pattern across VLM layers, we further introduce Budget-Deterministic Pruning (BDP) to establish a hardware-friendly schedule with fixed per-layer budgets. Extensive experimental results across various benchmarks and VLMs demonstrate that LearnPrunerV2 consistently outperform previous state-of-the-art methods. Notably, LearnPrunerV2 preserves 99.8% of the original performance when pruning 77.8% of visual tokens on LLaVA-v1.5, achieving a superior accuracy-efficiency trade-off.
Oct 26, 2026

TP-JSCC: Efficient Token Pruning and Joint Source-Channel Coding for LVLMs-based 3D-VQA in Cloud-Edge Collaboration

Deploying large vision-language models (LVLMs) on resource-constrained embodied agents for 3D visual question answering necessitates cloud-edge collaboration. However, offloading massive visual data to the cloud over bandwidth-limited and noisy channels introduces severe latency and transmission distortion. To address this, we propose TP-JSCC, an efficient token pruning and joint source-channel coding method for LVLMs-based 3D-VQA in Cloud-Edge Collaboration. First, we develop a Local and Global Semantics-based Hybrid Token Pruning (LGS-HTP) module to significantly reduce the visual token count. By decoupling foreground-background tokens and jointly selecting local and global semantics, LGS-HTP effectively reduces redundancy while preserving reasoning-critical tokens. Second, we introduce a lightweight Token Joint Source-Channel Coding (T-JSCC) module to further compress the pruned tokens and ensure their robust transmission over error-prone channels. By employing lightweight cascaded linear layers and Signal-to-Noise Ratio (SNR)-adaptive mechanism, this module adapts to fluctuating channel conditions while maintaining minimal computational overhead. Extensive experiments on ScanQA and SQA3D datasets indicate the exceptional efficiency and robustness of our approach. Specifically, on ScanQA validation set, the LGS-HTP module could achieve 90% pruning ratio while incurring less than a 1% CIDEr loss. Building on this, the T-JSCC further compresses the tokens by 18 times, resulting in an overall 180 times data volume reduction for robust transmission, while saving 90% in computational complexity for LVLM deployment. Our code is available at https://github.com/yuanhui0325/TP-JSCC.
Oct 25, 2026

VTC-AUDIT: An Interventional Reliability Audit for Visual-Token Compression under Paired Image Shifts

Visual-token compression is usually evaluated by clean accuracy and efficiency, but these averages do not reveal failures that compression introduces under an image shift where the uncompressed model stays correct. We introduce VTC-AUDIT, which estimates this incremental risk only on pairs where the full model is correct on both images and the compressed model is correct on the clean image. The four arms then cross two binary choices: which image selects the retained token positions, and which image supplies the visual features. We audit three open 8-9B vision-language models on four tasks and eight pre-specified shifts. We evaluate two training-free compressors---hard-pruning DivPrune and prune-and-merge ApET---at a matched 25% retained-position budget, producing 708,000 inference records. Reselecting positions on the shifted image, rather than reusing the clean-image positions, raises the failure rate in all six model-compressor settings. The increase is 3.04-9.60 percentage points for DivPrune and 2.05-4.20 percentage points for ApET, with every 95% interval above zero. Repeating the audit on a separately resampled second manifest reproduces all six increases. In a pooled monitor audit across the four benchmarks, support Jaccard instability fails the practical AUROC criterion of 0.65 in all six settings, whereas clean compressed-model uncertainty clears it. Thus, support turnover is a weak ranking signal under the audited benchmark mixture, while clean compressed-model uncertainty is substantially more informative. These results suggest that future visual-token compression evaluations should report conditional reliability alongside clean accuracy and efficiency.
Oct 25, 2026

TRIAGE: Efficient Vision-Language-Action Inference Needs Both Reuse and Pruning

Vision-language-action (VLA) models are increasingly used for robot manipulation because they map camera images and a language instruction directly to actions. Robots that cannot be tethered to a workstation, such as mobile manipulators and humanoids, must run them on onboard embedded boards, where the prefill latency of the language backbone grows with the hundreds of visual tokens processed at every control step. To meet real-time limits, training-free accelerators recompute only a subset of the visual tokens and skip the rest, either by pruning them or by reusing their cached key-value (KV) states. The two options save nearly the same computation, yet existing methods apply only one of them to every skipped token. Through a token-level diagnostic, we show that which option keeps the action closer to full recomputation depends on the token's drift, the change in its shallow-layer value since its cache was computed. On both backbones tested, reusing tokens below a drift threshold was no worse than pruning them, whereas beyond it, reuse yields up to 4.4x the action error of pruning; the same diagnostic calibrates this threshold per backbone. Building on this finding, we propose TRIAGE, which deems a cache valid when its drift is below this threshold and it is not too old. TRIAGE recomputes a fixed budget of tokens, prioritizing important tokens with invalid caches, reuses every other token with a valid cache, and prunes the rest; prune-only and reuse-only methods are the two extremes of this partition. Because the budget fixes the number of recomputed tokens, it directly sets the prefill latency. On LIBERO and SimplerEnv, TRIAGE stays within 3 pp of full recomputation in average success rate while cutting prefill latency on a Jetson AGX Orin by up to 48%, and with only 7% of visual tokens recomputed, it beats the strongest baseline by 24.4 pp on LIBERO-Object.
Oct 24, 2026

Rescuing Your Pruning: Rethinking Visual Token Pruning and Post-Training Recovery in MLLMs

Visual token pruning substantially reduces the inference cost of multimodal large language models (MLLMs), yet existing methods are mostly training-free and plug-and-play. They treat the resulting performance drop as an inevitable cost of compression and pursue increasingly complex pruning rules. We rethink this view through two questions: how much of the loss can post-training rescue, and, building on this recovery, how can joint compression push pruning to its limit? We curate a 282K-sample post-training mixture covering general VQA, reasoning, OCR, and grounding, and post-train 36 configurations that drop or merge tokens under a unified protocol. Post-training rescues most of the loss, even for OCR and grounding, and its adaptation gain in closing the gap between the pruned and the dense model grows as the budget tightens. For joint compression, we find that the vision encoder, not the LLM, dominates prefill time in native-resolution MLLMs, so where tokens are removed matters more than how many: keeping 40% of tokens at the encoder runs faster than keeping 10% at the projector. Our best configuration prunes the encoder lightly and the projector aggressively, leaves the LLM intact, and keeps general understanding and reasoning at the dense level. It accelerates the LLM stage by 3.49x but the end-to-end time-to-first-token by only 1.54x, showing that LLM-stage speedups overstate end-to-end gains. We hope these findings offer insights for future research on joint visual token compression. Our data, protocol, and code will be released.
Oct 23, 2026

Vision Token Pruning via Query-Vision Interaction Decomposition

Large vision-language models (VLMs), which process both visual inputs and text queries, incur high inference costs due to the large number of visual tokens, making visual token pruning a natural approach for improving efficiency. Recent work has leveraged query information to guide token selection, often through heuristics derived from attention patterns or token dynamics. However, these approaches do not explicitly exploit the underlying structure of query-vision interactions. We address this gap by observing that the query-vision interaction matrix, naturally induced by the query and key projections already present in transformer attention, is effectively low-rank, with a small number of dominant latent interaction modes capturing query-relevant semantics. Motivated by this observation, we propose Query-Vision Decomposition (QViD), a novel training-free query-aware visual token pruning method that exploits this structure. Extensive experiments on both image and video understanding benchmarks show that QViD consistently improves performance under matched token budgets, with the largest gains appearing under aggressive compression regimes.
Oct 20, 2026

GazeLens for Efficient Egocentric Video Question Answering with Simple Gaze-Driven Adaptive Token Reduction

Smart glasses can help users understand their activities and surroundings, but small vision-language models must preserve useful evidence within limited computation. This is difficult because redundant frames consume the visual budget, while aggressive spatial compression can discard relevant details. Although gaze identifies the wearer's focus, the question may require evidence from the surrounding scene. To address these constraints, we propose GazeLens, which combines frame selection with gaze- and question-guided token reduction. First, HOG transitions and image sharpness select frames before visual encoding. For each selected frame, Gaussian pooling then forms gaze-focused and complementary summaries. Because their relevance depends on the question, a lightweight gate blends them into a primary token. Together with a global scene token, this retains both focused and contextual summaries within two tokens per frame. On StreamGaze, GazeLens achieves higher macro and micro accuracy than the evaluated full-grid and token-reduction baselines, with lower input-forward FLOPs than the shared-backbone token-reduction baselines.
Oct 18, 2026

Visual Sufficiency Landscapes: Understanding Query-Conditioned Reliability in Visual Token Pruning

Different visual token pruning strategies retain different subsets while maintaining similar task performance. This motivates a combinatorial question: how many visual subsets can preserve model behavior, and how does their abundance depend on the query? We introduce visual sufficiency landscapes to characterize this abundance. At a fixed layer and budget, visual sufficiency probability is the fraction of uniformly sampled subsets that jointly preserve the dense model's answer, reference-answer likelihood, and output distribution. Independent sampling replicates separate input variation from estimation noise. Across Qwen2.5-VL-7B, InternVL2-8B, and LLaVA-v1.5-7B, differences between inputs on GQA and TextVQA exceed replication error, and different questions about the same GQA image produce distinct landscapes. Sufficiency probability also improves failure prediction for full pruning methods. On LLaVA, adding this probability increases AUC by 0.1151 for FastV, 0.0683 for VisionZip, and 0.1401 for SparseVLM, while reducing log loss and Brier score. Several architecture-adapted configurations on the other two models reproduce these gains. Under controlled selection rules, predictors trained on the original images reduce log loss on new images by 9.0%, 13.0%, and 14.1%, respectively. These results establish successful-subset abundance as a measurable dimension of local prunability and connect it to the identification of high-risk pruning conditions.
Oct 17, 2026

Are We Evaluating VLM Token Pruning Right? Looking Beyond Accuracy and Token Count

Visual token pruning aims to reduce inference cost while preserving the capabilities of the original VLM. Yet existing evaluations rely on aggregate task accuracy and retained token count as proxies for pruning quality and actual efficiency. We reveal two limitations: 1) Aggregate accuracy can obscure whether a pruning method truly identifies critical visual tokens; many examples remain solvable under arbitrary token removal and models can exploit textual shortcuts when visual evidence is degraded. 2) Retained token count is an unreliable proxy for efficiency, as identical token budgets can incur different end-to-end costs depending on where and how pruning is performed. Consequently, comparisons based on aggregate accuracy and retained token counts fail to recover the global performance-efficiency Pareto frontier across methods. We introduce PAA-Bench, a plug-and-play evaluation framework for existing benchmarks, to evaluate whether pruning succeeds through meaningful token selection and whether this success translates into real speedup. To determine when token choice actually matters, we propose Random-Pass@k to measure how often an example remains correct under random pruning, enabling comprehensive evaluation across different levels of selection difficulty. To compare efficiency fairly, we introduce Preservation-Constrained Speedup (PCS) to report the maximum end-to-end speedup each method achieves under a shared preservation target. Experiments across 8 methods, 3 VLM backbones, and 9 benchmarks show that stronger token compression does not necessarily yield greater speedup, method differences widen on examples where token selection matters, and efficiency rankings change with the preservation target.