OVD Object Detection Using Captions, CVPR 2021 Oral

1. Motivation

However, though deep learning models exhibit outstanding performance in object detection tasks, they are expensive to train and deploy due to supervisory demands.

Weakly supervised and zero-shot learning techniques have been investigated to extend object detectors to more categories with fewer supervision cues, but they have not achieved the same level of success or have not been as widely accepted as supervised models.

For addressing the challenge of OVD