What Makes for Good Visual Tokenizers for Large Language Models?

TL;DR

This paper empirically evaluates various visual tokenizers for large multimodal models, revealing insights on their semantic and fine-grained perception capabilities, and introduces a new tokenizer that enhances visual understanding without extra parameters.

Contribution

It provides a comprehensive benchmark of visual tokenizers, analyzes their strengths and weaknesses, and proposes a tailored Good Visual Tokenizer (GVT) that improves multimodal model performance.

Findings

Fully/weakly supervised models capture more semantics than self-supervised models.

Scaling pre-training datasets narrows the semantic gap.

Self-supervised models excel at fine-grained perception.

Abstract

We empirically investigate proper pre-training methods to build good visual tokenizers, making Large Language Models (LLMs) powerful Multimodal Large Language Models (MLLMs). In our benchmark, which is curated to evaluate MLLMs visual semantic understanding and fine-grained perception capabilities, we discussed different visual tokenizers pre-trained with dominant methods (i.e., DeiT, CLIP, MAE, DINO), and observe that: i) Fully/weakly supervised models capture more semantics than self-supervised models, but the gap is narrowed by scaling up the pre-training dataset. ii) Self-supervised models are better at fine-grained perception, where patch-level supervision is particularly effective. iii) Tuning the visual tokenizer leads to the loss of semantics obtained from large-scale pretraining, which is unfavorable with relatively small-scale instruction-tuning dataset. Given the findings, we reviewed methods that attempted to unify semantics and fine-grained visual understanding, e.g., patch-level feature distillation with semantically-rich targets. We obtain an intriguing insight mask-based strategies that were once all the rage may not be applicable for obtaining good visual tokenizers. Based on this critical observation, we obtain a new MLLM equipped with a tailored Good Visual Tokenizer (GVT), which exhibits strong visual comprehension capability at multiple scales. In particular, without introducing extra parameters and task-specific fine-tuning, GVT achieves superior performance on visual question answering, image captioning, and other fine-grained visual understanding tasks such as object counting and multi-class identification.