VoCo-LLaMA: Towards Vision Compression with Large Language Models

2024/06/18 by Xubing Ye, Ye, Xubing, Yukang Gan +7 · 15 citations
Computer Science · #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Image Retrieval and Classification Techniques #Multimodal Machine Learning Applications

paper · pdf · doi:10.48550/arxiv.2406.12275

openalex publication_date 2024/06/18 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Vision-Language Models (VLMs) have achieved remarkable success in various multi-modal tasks, but they are often bottlenecked by the limited context window and high computational cost of processing high-resolution image inputs and videos. Vision compression can alleviate this problem by reducing the vision token count. Previous approaches compress vision tokens with external modules and force LLMs to understand the compressed ones, leading to visual information loss. However, the LLMs' understanding paradigm of vision tokens is not fully utilised in the compression learning process. We propose VoCo-LLaMA, the first approach to compress vision tokens using LLMs. By introducing Vision Compression tokens during the vision instruction tuning phase and leveraging attention distillation, our method distill how LLMs comprehend vision tokens into their processing of VoCo tokens. VoCo-LLaMA facilitates effective vision compression and improves the computational efficiency during the inference stage. Specifically, our method achieves minimal performance loss with a compression ratio of 576×, resulting in up to 94.8% fewer FLOPs and 69.6% acceleration in inference time. Furthermore, through continuous training using time-series compressed token sequences of video frames, VoCo-LLaMA demonstrates the ability to understand temporal correlations, outperforming previous methods on popular video question-answering benchmarks. Our approach presents a promising way to unlock the full potential of VLMs' contextual window, enabling more scalable multi-modal applications. The project page, along with the associated code, can be accessed via https://yxxxb.github.io/VoCo-LLaMA-page/.

Cited by

• AstraNav-Memory: Contexts Compression for Long Memory
• Widget2Code: From Visual Widgets to UI Code via Multimodal LLMs
• Adaptive-VoCo: Complexity-Aware Visual Token Compression for Vision-Language Models
• Focus: A Streaming Concentration Architecture for Efficient Vision-Language Models
• HybridToken-VLM: Hybrid Token Compression for Vision-Language Models
• Parallel Vision Token Scheduling for Fast and Accurate Multimodal LMMs Inference
• VisReason: A Large-Scale Dataset for Visual Chain-of-Thought Reasoning
• TimeViper: A Hybrid Mamba-Transformer Vision-Language Model for Efficient Long Video Understanding
• Auto-scaling Continuous Memory for GUI Agent
• Efficient Multi-modal Large Language Models via Progressive Consistency Distillation
• FrameMind: Frame-Interleaved Video Reasoning via Reinforcement Learning
• Energy-Driven Adaptive Visual Token Pruning for Efficient Vision-Language Models
• SpecVLM: Fast Speculative Decoding in Vision-Language Models
• Quantization Meets dLLMs: A Systematic Study of Post-training Quantization for Diffusion LLMs
• METEOR: Multi-Encoder Collaborative Token Pruning for Efficient Vision Language Models

Related

• Adaptive-VoCo: Complexity-Aware Visual Token Compression for Vision-Language Models
• LLMC+: Benchmarking Vision-Language Model Compression with a Plug-and-play Toolkit
• IPCV: Information-Preserving Compression for MLLM Visual Encoders
• CS-VLM: Compressed Sensing Attention for Efficient Vision-Language Representation Learning
• Variation-aware Vision Token Dropping for Faster Large Vision-Language Models