GLIPv2: Unifying Localization and Vision-Language Understanding

TL;DR

GLIPv2 is a unified vision-language model that combines localization and understanding tasks through novel pre-training tasks, achieving near state-of-the-art results and strong zero-shot capabilities.

Contribution

It introduces a unified pre-training framework that integrates localization and vision-language understanding tasks, simplifying previous methods and enhancing performance.

Findings

Achieves near state-of-the-art on multiple tasks

Demonstrates strong zero-shot and few-shot detection

Shows superior grounding capabilities

Abstract

We present GLIPv2, a grounded VL understanding model, that serves both localization tasks (e.g., object detection, instance segmentation) and Vision-Language (VL) understanding tasks (e.g., VQA, image captioning). GLIPv2 elegantly unifies localization pre-training and Vision-Language Pre-training (VLP) with three pre-training tasks: phrase grounding as a VL reformulation of the detection task, region-word contrastive learning as a novel region-word level contrastive learning task, and the masked language modeling. This unification not only simplifies the previous multi-stage VLP procedure but also achieves mutual benefits between localization and understanding tasks. Experimental results show that a single GLIPv2 model (all model weights are shared) achieves near SoTA performance on various localization and understanding tasks. The model also shows (1) strong zero-shot and few-shot adaption performance on open-vocabulary object detection tasks and (2) superior grounding capability on VL understanding tasks. Code will be released at https://github.com/microsoft/GLIP.