Visual Token Pruning
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.
ST-SimDiff: Balancing Spatiotemporal Similarity and Difference for Efficient Video Understanding with MLLMs
Rethinking Token Reduction for Diffusion Models via Output-Similarity-Awareness
Focus-then-Context: Subject-Centric Progressive Visual Token Reduction for Vision-Language Models
Faster or Stronger: Towards Flexible Visual Place Recognition via Weighted Aggregation and Token Pruning
EventPrune: Cascaded Event-Assisted Token Pruning for Efficient First-Person Dynamic Spatial Reasoning
DynaTok: Temporally Adaptive and Positional Bias-Aware Token Compression for Video-LLMs
Temporal Aware Pruning for Efficient Diffusion-based Video Generation
FastOCR: Dynamic Visual Fixation via KV Cache Pruning for Efficient Document Parsing
GHOST: Geometry-Hierarchical Online Streaming Token Eviction for Efficient 3D Reconstruction
LRCP: Low-Rank Compressibility Guided Visual Token Pruning for Efficient LVLMs
CoReDiT: Spatial Coherence-Guided Token Pruning and Reconstruction for Efficient Diffusion Transformers
GRIP-VLM: Group-Relative Importance Pruning for Efficient Vision-Language Models
CLIP Tricks You: Training-free Token Pruning for Efficient Pixel Grounding in Large VIsion-Language Models
Learning to See What You Need: Gaze Attention for Multimodal Large Language Models
Keep What Audio Cannot Say: Context-Preserving Token Pruning for Omni-LLMs
LDDR: Linear-DPP Dynamic-Resolution Frame Sampling for Video MLLMs
ReVision: Scaling Computer-Use Agents via Temporal Visual Redundancy Reduction
Mutual Enhancement Between Global Tokens and Patch Tokens: From Theory to Practice
EchoPrune: Interpreting Redundancy as Temporal Echoes for Efficient VideoLLMs
ERASE: Eliminating Redundant Visual Tokens via Adaptive Two-Stage Token Pruning
Evading Visual Aphasia: Contrastive Adaptive Semantic Token Pruning for Vision-Language Models
How Many Visual Tokens Do Multimodal Language Models Need? Scaling Visual Token Pruning with F^3A
PrePARE: Pre-AA Token Pruning for Frozen Multi-View Geometry Transformers
An Efficient Token Compression Framework for Visual Object Tracking
LookWhen? Fast Video Recognition by Learning When, Where, and What to Compute
Spark3R: Asymmetric Token Reduction Makes Fast Feed-Forward 3D Reconstruction
LIVEditor-14B: Lightning Unified Video Editing via In-Context Sparse Attention
LIVEditor-14B , a novel lightning video editing model via ISA and a proposed video-editing data pipeline that curated a 1.7M high-quality dataset. Extensive experiments demonstrate that LIVEditor-14B achieves a 60% reduction in attention-module latency while surpassing state-of-the-art methods across EditVerseBench, IVE-Bench, and VIE-Bench, delivering near-lossless acceleration without compromising visual fidelity.