Towards Semantic Equivalence of Tokenization in Multimodal LLM

TL;DR

This paper introduces SeTok, a dynamic vision tokenizer for multimodal large language models that preserves semantic integrity of visual features, improving performance across vision-language tasks.

Contribution

The paper proposes a novel dynamic clustering-based vision tokenizer, SeTok, which better preserves semantic information compared to existing methods.

Findings

SeTok effectively maintains semantic integrity of visual features.

Setokim with SeTok outperforms existing models on multiple tasks.

The dynamic clustering adapts to image complexity for optimal tokenization.

Abstract

Multimodal Large Language Models (MLLMs) have demonstrated exceptional capabilities in processing vision-language tasks. One of the crux of MLLMs lies in vision tokenization, which involves efficiently transforming input visual signals into feature representations that are most beneficial for LLMs. However, existing vision tokenizers, essential for semantic alignment between vision and language, remain problematic. Existing methods aggressively fragment visual input, corrupting the visual semantic integrity. To address this, this paper proposes a novel dynamic Semantic-Equivalent Vision Tokenizer (SeTok), which groups visual features into semantic units via a dynamic clustering algorithm, flexibly determining the number of tokens based on image complexity. The resulting vision tokens effectively preserve semantic integrity and capture both low-frequency and high-frequency visual features. The proposed MLLM (Setokim) equipped with SeTok significantly demonstrates superior performance across various tasks, as evidenced by our experimental results. The project page is at https://chocowu.github.io/SeTok-web/.