InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning

TL;DR

InstructBLIP is a vision-language model trained with instruction tuning on diverse datasets, achieving state-of-the-art zero-shot and fine-tuned performance, advancing general-purpose multimodal AI capabilities.

Contribution

This work systematically studies vision-language instruction tuning, introduces an instruction-aware Query Transformer, and demonstrates significant performance improvements over existing models.

Findings

State-of-the-art zero-shot performance on 13 datasets

High accuracy on downstream tasks like ScienceQA (90.7%)

Models outperform BLIP-2 and Flamingo in various evaluations

Abstract

Large-scale pre-training and instruction tuning have been successful at creating general-purpose language models with broad competence. However, building general-purpose vision-language models is challenging due to the rich input distributions and task diversity resulting from the additional visual input. Although vision-language pretraining has been widely studied, vision-language instruction tuning remains under-explored. In this paper, we conduct a systematic and comprehensive study on vision-language instruction tuning based on the pretrained BLIP-2 models. We gather 26 publicly available datasets, covering a wide variety of tasks and capabilities, and transform them into instruction tuning format. Additionally, we introduce an instruction-aware Query Transformer, which extracts informative features tailored to the given instruction. Trained on 13 held-in datasets, InstructBLIP attains state-of-the-art zero-shot performance across all 13 held-out datasets, substantially outperforming BLIP-2 and larger Flamingo models. Our models also lead to state-of-the-art performance when finetuned on individual downstream tasks (e.g., 90.7% accuracy on ScienceQA questions with image contexts). Furthermore, we qualitatively demonstrate the advantages of InstructBLIP over concurrent multimodal models. All InstructBLIP models are open-sourced at https://github.com/salesforce/LAVIS/tree/main/projects/instructblip.