LLaVA-UHD: an LMM Perceiving Any Aspect Ratio and High-Resolution Images

TL;DR

LLaVA-UHD is a large multimodal model capable of perceiving images at any aspect ratio and high resolution efficiently, overcoming limitations of fixed-size processing in existing models.

Contribution

The paper introduces LLaVA-UHD with a novel image modularization, compression, and spatial organization strategy, enabling high-resolution, aspect ratio flexible image understanding in multimodal models.

Findings

Outperforms larger models on 9 benchmarks

Supports 6 times larger images with less computation

Achieves 6.4% improvement on TextVQA

Abstract

Visual encoding constitutes the basis of large multimodal models (LMMs) in understanding the visual world. Conventional LMMs process images in fixed sizes and limited resolutions, while recent explorations in this direction are limited in adaptivity, efficiency, and even correctness. In this work, we first take GPT-4V and LLaVA-1.5 as representative examples and expose systematic flaws rooted in their visual encoding strategy. To address the challenges, we present LLaVA-UHD, a large multimodal model that can efficiently perceive images in any aspect ratio and high resolution. LLaVA-UHD includes three key components: (1) An image modularization strategy that divides native-resolution images into smaller variable-sized slices for efficient and extensible encoding, (2) a compression module that further condenses image tokens from visual encoders, and (3) a spatial schema to organize slice tokens for LLMs. Comprehensive experiments show that LLaVA-UHD outperforms established LMMs trained with 2-3 orders of magnitude more data on 9 benchmarks. Notably, our model built on LLaVA-1.5 336x336 supports 6 times larger (i.e., 672x1088) resolution images using only 94% inference computation, and achieves 6.4 accuracy improvement on TextVQA. Moreover, the model can be efficiently trained in academic settings, within 23 hours on 8 A100 GPUs (vs. 26 hours of LLaVA-1.5). We make the data and code publicly available at https://github.com/thunlp/LLaVA-UHD.