ViCToR: Improving Visual Comprehension via Token Reconstruction for Pretraining LMMs

TL;DR

ViCToR introduces a novel pretraining framework for large multimodal models that enhances visual understanding by reconstructing visual tokens, leading to state-of-the-art performance on multiple benchmarks.

Contribution

The paper proposes ViCToR, a new pretraining method that improves visual token representation in LMMs using token reconstruction and semantic supervision.

Findings

Achieves state-of-the-art results on MMStar, SEED$^I$, and RealWorldQA benchmarks.

Improves LLaVA-NeXT-8B performance by over 10% on key benchmarks.

Demonstrates the effectiveness of visual token reconstruction in LMM pretraining.

Abstract

Large Multimodal Models (LMMs) often face a modality representation gap during pretraining: while language embeddings remain stable, visual representations are highly sensitive to contextual noise (e.g., background clutter). To address this issue, we introduce a visual comprehension stage, which we call ViCToR (Visual Comprehension via Token Reconstruction), a novel pretraining framework for LMMs. ViCToR employs a learnable visual token pool and utilizes the Hungarian matching algorithm to select semantically relevant tokens from this pool for visual token replacement. Furthermore, by integrating a visual token reconstruction loss with dense semantic supervision, ViCToR can learn tokens which retain high visual detail, thereby enhancing the large language model's (LLM's) understanding of visual information. After pretraining on 3 million publicly accessible images and captions, ViCToR achieves state-of-the-art results, improving over LLaVA-NeXT-8B by 10.4%, 3.2%, and 7.2% on the MMStar, SEED$^I$, and RealWorldQA benchmarks, respectively. Code is available at https://github.com/deepglint/Victor.