Scaling Open-Vocabulary Object Detection

TL;DR

This paper introduces OWLv2 and OWL-ST, scalable methods for open-vocabulary object detection using web data, significantly improving performance on rare classes by leveraging self-training at unprecedented scales.

Contribution

The paper presents the OWLv2 model and OWL-ST self-training method, enabling scalable open-vocabulary detection with large-scale web data, surpassing previous state-of-the-art results.

Findings

OWLv2 outperforms previous detectors at similar training scales.

OWL-ST scales training to over 1 billion examples, greatly improving rare class detection.

AP on LVIS rare classes increased from 31.2% to 44.6%.

Abstract

Open-vocabulary object detection has benefited greatly from pretrained vision-language models, but is still limited by the amount of available detection training data. While detection training data can be expanded by using Web image-text pairs as weak supervision, this has not been done at scales comparable to image-level pretraining. Here, we scale up detection data with self-training, which uses an existing detector to generate pseudo-box annotations on image-text pairs. Major challenges in scaling self-training are the choice of label space, pseudo-annotation filtering, and training efficiency. We present the OWLv2 model and OWL-ST self-training recipe, which address these challenges. OWLv2 surpasses the performance of previous state-of-the-art open-vocabulary detectors already at comparable training scales (~10M examples). However, with OWL-ST, we can scale to over 1B examples, yielding further large improvement: With an L/14 architecture, OWL-ST improves AP on LVIS rare classes, for which the model has seen no human box annotations, from 31.2% to 44.6% (43% relative improvement). OWL-ST unlocks Web-scale training for open-world localization, similar to what has been seen for image classification and language modelling.