Few-shot target-driven instance detection based on open-vocabulary object detection models

TL;DR

This paper introduces a lightweight method leveraging open-vocabulary object detection models for one-shot and few-shot target-driven instance detection, avoiding costly re-training and enabling efficient recognition.

Contribution

The authors propose a novel approach to adapt open-vocabulary detection models for few-shot detection without textual descriptions, improving efficiency and flexibility.

Findings

Performance improves with larger models

Adding more examples enhances detection accuracy

Image augmentation boosts detection performance

Abstract

Current large open vision models could be useful for one and few-shot object recognition. Nevertheless, gradient-based re-training solutions are costly. On the other hand, open-vocabulary object detection models bring closer visual and textual concepts in the same latent space, allowing zero-shot detection via prompting at small computational cost. We propose a lightweight method to turn the latter into a one-shot or few-shot object recognition models without requiring textual descriptions. Our experiments on the TEgO dataset using the YOLO-World model as a base show that performance increases with the model size, the number of examples and the use of image augmentation.