Osprey: Pixel Understanding with Visual Instruction Tuning

TL;DR

Osprey introduces a mask-text instruction tuning method for vision-language models, enabling pixel-level understanding by incorporating fine-grained mask regions and a large mask-based dataset, enhancing region understanding capabilities.

Contribution

The paper presents a novel mask-text instruction tuning approach and a large dataset, advancing pixel-wise visual understanding in multimodal large language models.

Findings

Osprey outperforms existing models in region understanding tasks.

It can be integrated with SAM for multi-granularity semantics.

Demonstrates effective fine-grained pixel-level visual understanding.

Abstract

Multimodal large language models (MLLMs) have recently achieved impressive general-purpose vision-language capabilities through visual instruction tuning. However, current MLLMs primarily focus on image-level or box-level understanding, falling short in achieving fine-grained vision-language alignment at pixel level. Besides, the lack of mask-based instruction data limits their advancements. In this paper, we propose Osprey, a mask-text instruction tuning approach, to extend MLLMs by incorporating fine-grained mask regions into language instruction, aiming at achieving pixel-wise visual understanding. To achieve this goal, we first meticulously curate a mask-based region-text dataset with 724K samples, and then design a vision-language model by injecting pixel-level representation into LLM. Specifically, Osprey adopts a convolutional CLIP backbone as the vision encoder and employs a mask-aware visual extractor to extract precise visual mask features from high resolution input. Experimental results demonstrate Osprey's superiority in various region understanding tasks, showcasing its new capability for pixel-level instruction tuning. In particular, Osprey can be integrated with Segment Anything Model (SAM) seamlessly to obtain multi-granularity semantics. The source code, dataset and demo can be found at https://github.com/CircleRadon/Osprey.