Reroute turns irreversible visual-token pruning into recoverable routing that reuses existing attention scores, improving grounding performance under aggressive reduction on LLaVA-1.5 and Qwen while preserving TFLOPs and KV-cache budgets.
Multi-stage vision token dropping: Towards efficient multimodal large language model
12 Pith papers cite this work. Polarity classification is still indexing.
12
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
roles
background 1polarities
background 1representative citing papers
RotateK uses online PCA-based rotation to align token-dependent key channel importance into a shared subspace, enabling accurate head-wise structured pruning and faster decoding in VLMs compared to prior token or channel methods.
CAGE attack aligns perturbations with token compression to achieve lower robust accuracy on compressed LVLMs than baseline attacks across mechanisms and datasets.
MS-Resampler deploys multiple scope-specific resamplers with explicit spatial priors and adaptive fusion to outperform single-scope global cross-attention in MLLMs on ten benchmarks with minimal added cost.
TOPS formulates visual token pruning as constructing Token Optimal Preservation Sets using three information-theoretic principles and demonstrates superior performance on MLLM benchmarks.
PARCEL is a new visual tokenization architecture combining pool-anchored resampling with conditioned elastic queries to enhance performance-efficiency tradeoffs in LVLMs over prior matryoshka methods.
EarlyTom is a training-free early token compression method inside the vision encoder with decoupled spatial selection that reduces TTFT up to 2.65x and FLOPs 61% on LLaVA-OneVision-7B while keeping accuracy comparable to full tokens.
POINTS-Long is a dual-mode multimodal large language model that uses dynamic visual token scaling to retain 97.7-99.7% accuracy on long-form tasks with 1/40 to 1/10th the tokens and supports streaming via detachable KV-cache.
FastOCR dynamically selects a small subset of visual tokens per decoding step using focal-guided pruning and cross-step reuse, retaining 98% accuracy on Qwen2.5-VL while attending to only 5% of tokens and cutting attention latency by 3x.
EvoComp compresses visual tokens in MLLMs by 3x while retaining 99.3% accuracy via an evolutionary labeling strategy that searches for low-loss, semantically diverse token subsets.
AOT reduces visual tokens in VLLMs via intra-frame and inter-frame anchors with local-global optimal transport, delivering competitive benchmark performance and efficiency gains in a training-free way.
TwigVLM adds a twig module to VLMs for twig-guided token pruning and self-speculative decoding, retaining 96% performance after pruning 88.9% visual tokens and delivering 154% speedup on long responses for LLaVA-1.5-7B.
citing papers explorer
-
Reroute, Don't Remove: Recoverable Visual Token Routing for Vision-Language Models
Reroute turns irreversible visual-token pruning into recoverable routing that reuses existing attention scores, improving grounding performance under aggressive reduction on LLaVA-1.5 and Qwen while preserving TFLOPs and KV-cache budgets.
-
Rotation-Aligned Key Channel Pruning for Efficient Vision-Language Model Inference
RotateK uses online PCA-based rotation to align token-dependent key channel importance into a shared subspace, enabling accurate head-wise structured pruning and faster decoding in VLMs compared to prior token or channel methods.
-
On the Adversarial Robustness of Large Vision-Language Models under Visual Token Compression
CAGE attack aligns perturbations with token compression to achieve lower robust accuracy on compressed LVLMs than baseline attacks across mechanisms and datasets.
-
MS-Resampler: Multi-Scope Visual Resampling for Efficient Multimodal LLMs
MS-Resampler deploys multiple scope-specific resamplers with explicit spatial priors and adaptive fusion to outperform single-scope global cross-attention in MLLMs on ten benchmarks with minimal added cost.
-
TOPS: First-Principles Visual Token Pruning via Constructing Token Optimal Preservation Sets for Efficient MLLM Inference
TOPS formulates visual token pruning as constructing Token Optimal Preservation Sets using three information-theoretic principles and demonstrates superior performance on MLLM benchmarks.
-
PARCEL: Pool-Anchored Resampling with Conditioned Elastic Queries for Efficient Vision-Language Understanding
PARCEL is a new visual tokenization architecture combining pool-anchored resampling with conditioned elastic queries to enhance performance-efficiency tradeoffs in LVLMs over prior matryoshka methods.
-
EarlyTom: Early Token Compression Completes Fast Video Understanding
EarlyTom is a training-free early token compression method inside the vision encoder with decoupled spatial selection that reduces TTFT up to 2.65x and FLOPs 61% on LLaVA-OneVision-7B while keeping accuracy comparable to full tokens.
-
POINTS-Long: Adaptive Dual-Mode Visual Reasoning in MLLMs
POINTS-Long is a dual-mode multimodal large language model that uses dynamic visual token scaling to retain 97.7-99.7% accuracy on long-form tasks with 1/40 to 1/10th the tokens and supports streaming via detachable KV-cache.
-
FastOCR: Dynamic Visual Fixation via KV Cache Pruning for Efficient Document Parsing
FastOCR dynamically selects a small subset of visual tokens per decoding step using focal-guided pruning and cross-step reuse, retaining 98% accuracy on Qwen2.5-VL while attending to only 5% of tokens and cutting attention latency by 3x.
-
EvoComp: Learning Visual Token Compression for Multimodal Large Language Models via Semantic-Guided Evolutionary Labeling
EvoComp compresses visual tokens in MLLMs by 3x while retaining 99.3% accuracy via an evolutionary labeling strategy that searches for low-loss, semantically diverse token subsets.
-
Token Reduction via Local and Global Contexts Optimization for Efficient Video Large Language Models
AOT reduces visual tokens in VLLMs via intra-frame and inter-frame anchors with local-global optimal transport, delivering competitive benchmark performance and efficiency gains in a training-free way.
-
Growing a Multi-head Twig via Distillation and Reinforcement Learning to Accelerate Large Vision-Language Models
TwigVLM adds a twig module to VLMs for twig-guided token pruning and self-speculative decoding, retaining 96% performance after pruning 88.9% visual tokens and delivering 154% speedup on long responses for LLaVA-1.5-7B.