Simple Open-Vocabulary Object Detection with Vision Transformers

TL;DR

This paper presents a simple yet effective approach for open-vocabulary object detection using Vision Transformers, leveraging large-scale image-text pre-training and minimal architecture modifications to achieve strong zero-shot and one-shot detection performance.

Contribution

It introduces a straightforward method to adapt image-text models for open-vocabulary detection with minimal changes, emphasizing the importance of scaling and pre-training.

Findings

Scaling pre-training improves detection performance

Minimal architecture modifications are effective

Achieves strong zero-shot and one-shot detection results

Abstract

Combining simple architectures with large-scale pre-training has led to massive improvements in image classification. For object detection, pre-training and scaling approaches are less well established, especially in the long-tailed and open-vocabulary setting, where training data is relatively scarce. In this paper, we propose a strong recipe for transferring image-text models to open-vocabulary object detection. We use a standard Vision Transformer architecture with minimal modifications, contrastive image-text pre-training, and end-to-end detection fine-tuning. Our analysis of the scaling properties of this setup shows that increasing image-level pre-training and model size yield consistent improvements on the downstream detection task. We provide the adaptation strategies and regularizations needed to attain very strong performance on zero-shot text-conditioned and one-shot image-conditioned object detection. Code and models are available on GitHub.