PANDAS: Prototype-based Novel Class Discovery and Detection

Tyler L. Hayes, César Roberto de Souza, Namil Kim, Jiwon Kim, Riccardo Volpi, Diane Larlus
Proceedings of The 3rd Conference on Lifelong Learning Agents, PMLR 274:367-387, 2025.

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.

Cite this Paper

BibTeX
@InProceedings{pmlr-v274-hayes25a, title = {PANDAS: Prototype-based Novel Class Discovery and Detection}, author = {Hayes, Tyler L. and Souza, C{\'{e}}sar Roberto de and Kim, Namil and Kim, Jiwon and Volpi, Riccardo and Larlus, Diane}, booktitle = {Proceedings of The 3rd Conference on Lifelong Learning Agents}, pages = {367--387}, year = {2025}, editor = {Lomonaco, Vincenzo and Melacci, Stefano and Tuytelaars, Tinne and Chandar, Sarath and Pascanu, Razvan}, volume = {274}, series = {Proceedings of Machine Learning Research}, month = {29 Jul--01 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v274/main/assets/hayes25a/hayes25a.pdf}, url = {https://proceedings.mlr.press/v274/hayes25a.html}, 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.} }
Endnote
%0 Conference Paper %T PANDAS: Prototype-based Novel Class Discovery and Detection %A Tyler L. Hayes %A César Roberto de Souza %A Namil Kim %A Jiwon Kim %A Riccardo Volpi %A Diane Larlus %B Proceedings of The 3rd Conference on Lifelong Learning Agents %C Proceedings of Machine Learning Research %D 2025 %E Vincenzo Lomonaco %E Stefano Melacci %E Tinne Tuytelaars %E Sarath Chandar %E Razvan Pascanu %F pmlr-v274-hayes25a %I PMLR %P 367--387 %U https://proceedings.mlr.press/v274/hayes25a.html %V 274 %X 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.
APA
Hayes, T.L., Souza, C.R.d., Kim, N., Kim, J., Volpi, R. & Larlus, D.. (2025). PANDAS: Prototype-based Novel Class Discovery and Detection. Proceedings of The 3rd Conference on Lifelong Learning Agents, in Proceedings of Machine Learning Research 274:367-387 Available from https://proceedings.mlr.press/v274/hayes25a.html.

Related Material