Informed Mixing -- Improving Open Set Recognition via Attribution-based Augmentation
Fuente:
arXiv
Salvato in:
| Autori principali: | , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866910952965799936 |
|---|---|
| author | Xu, Jiawen Kao, Odej Keuper, Margret |
| author_facet | Xu, Jiawen Kao, Odej Keuper, Margret |
| contents | Open set recognition (OSR) is devised to address the problem of detecting novel classes during model inference. Even in recent vision models, this remains an open issue which is receiving increasing attention. Thereby, a crucial challenge is to learn features that are relevant for unseen categories from given data, for which these features might not be discriminative. To facilitate this process and "optimize to learn" more diverse features, we propose GradMix, a data augmentation method that dynamically leverages gradient-based attribution maps of the model during training to mask out already learned concepts. Thus GradMix encourages the model to learn a more complete set of representative features from the same data source. Extensive experiments on open set recognition, close set classification, and out-of-distribution detection reveal that our method can often outperform the state-of-the-art. GradMix can further increase model robustness to corruptions as well as downstream classification performance for self-supervised learning, indicating its benefit for model generalization. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_12803 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Informed Mixing -- Improving Open Set Recognition via Attribution-based Augmentation Xu, Jiawen Kao, Odej Keuper, Margret Computer Vision and Pattern Recognition Machine Learning Open set recognition (OSR) is devised to address the problem of detecting novel classes during model inference. Even in recent vision models, this remains an open issue which is receiving increasing attention. Thereby, a crucial challenge is to learn features that are relevant for unseen categories from given data, for which these features might not be discriminative. To facilitate this process and "optimize to learn" more diverse features, we propose GradMix, a data augmentation method that dynamically leverages gradient-based attribution maps of the model during training to mask out already learned concepts. Thus GradMix encourages the model to learn a more complete set of representative features from the same data source. Extensive experiments on open set recognition, close set classification, and out-of-distribution detection reveal that our method can often outperform the state-of-the-art. GradMix can further increase model robustness to corruptions as well as downstream classification performance for self-supervised learning, indicating its benefit for model generalization. |
| title | Informed Mixing -- Improving Open Set Recognition via Attribution-based Augmentation |
| topic | Computer Vision and Pattern Recognition Machine Learning |
| url | https://arxiv.org/abs/2505.12803 |