PANDAS: Prototype-based Novel Class Discovery and Detection

TL;DR

PANDAS is a simple, prototype-based method that extends object detectors to identify and learn new classes from unlabeled data, improving detection capabilities in dynamic environments.

Contribution

It introduces a novel prototype-based approach for joint discovery and detection of new classes in object detection tasks.

Findings

Performs favorably against state-of-the-art methods

More computationally affordable

Effective on VOC 2012 and COCO-to-LVIS benchmarks

Abstract

Object detectors are typically trained once and for all on a fixed set of classes. However, this closed-world assumption is unrealistic in practice, as new classes will inevitably emerge after the detector is deployed in the wild. In this work, we look at ways to extend a detector trained for a set of base classes so it can i) spot the presence of novel classes, and ii) automatically enrich its repertoire to be able to detect those newly discovered classes together with the base ones. We propose PANDAS, a method for novel class discovery and detection. It discovers clusters representing novel classes from unlabeled data, and represents old and new classes with prototypes. During inference, a distance-based classifier uses these prototypes to assign a label to each detected object instance. The simplicity of our method makes it widely applicable. We experimentally demonstrate the effectiveness of PANDAS on the VOC 2012 and COCO-to-LVIS benchmarks. It performs favorably against the state of the art for this task while being computationally more affordable.