A Contrastive Teacher-Student Framework for Novelty Detection under Style Shifts

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Mirzaei, Hossein, Nafez, Mojtaba, Madadi, Moein, Maleki, Arad, Hajialilue, Mahdi, Taghavi, Zeinab Sadat, Rezaee, Sepehr, Ansari, Ali, Nia, Bahar Dibaei, Shamsaie, Kian, Salehi, Mohammadreza, Mathis, Mackenzie W., Baghshah, Mahdieh Soleymani, Sabokrou, Mohammad, Rohban, Mohammad Hossein
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929690443251712
author Mirzaei, Hossein
Nafez, Mojtaba
Madadi, Moein
Maleki, Arad
Hajialilue, Mahdi
Taghavi, Zeinab Sadat
Rezaee, Sepehr
Ansari, Ali
Nia, Bahar Dibaei
Shamsaie, Kian
Salehi, Mohammadreza
Mathis, Mackenzie W.
Baghshah, Mahdieh Soleymani
Sabokrou, Mohammad
Rohban, Mohammad Hossein
author_facet Mirzaei, Hossein
Nafez, Mojtaba
Madadi, Moein
Maleki, Arad
Hajialilue, Mahdi
Taghavi, Zeinab Sadat
Rezaee, Sepehr
Ansari, Ali
Nia, Bahar Dibaei
Shamsaie, Kian
Salehi, Mohammadreza
Mathis, Mackenzie W.
Baghshah, Mahdieh Soleymani
Sabokrou, Mohammad
Rohban, Mohammad Hossein
contents There have been several efforts to improve Novelty Detection (ND) performance. However, ND methods often suffer significant performance drops under minor distribution shifts caused by changes in the environment, known as style shifts. This challenge arises from the ND setup, where the absence of out-of-distribution (OOD) samples during training causes the detector to be biased toward the dominant style features in the in-distribution (ID) data. As a result, the model mistakenly learns to correlate style with core features, using this shortcut for detection. Robust ND is crucial for real-world applications like autonomous driving and medical imaging, where test samples may have different styles than the training data. Motivated by this, we propose a robust ND method that crafts an auxiliary OOD set with style features similar to the ID set but with different core features. Then, a task-based knowledge distillation strategy is utilized to distinguish core features from style features and help our model rely on core features for discriminating crafted OOD and ID sets. We verified the effectiveness of our method through extensive experimental evaluations on several datasets, including synthetic and real-world benchmarks, against nine different ND methods.
format Preprint
id arxiv_https___arxiv_org_abs_2501_17289
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A Contrastive Teacher-Student Framework for Novelty Detection under Style Shifts
Mirzaei, Hossein
Nafez, Mojtaba
Madadi, Moein
Maleki, Arad
Hajialilue, Mahdi
Taghavi, Zeinab Sadat
Rezaee, Sepehr
Ansari, Ali
Nia, Bahar Dibaei
Shamsaie, Kian
Salehi, Mohammadreza
Mathis, Mackenzie W.
Baghshah, Mahdieh Soleymani
Sabokrou, Mohammad
Rohban, Mohammad Hossein
Computer Vision and Pattern Recognition
There have been several efforts to improve Novelty Detection (ND) performance. However, ND methods often suffer significant performance drops under minor distribution shifts caused by changes in the environment, known as style shifts. This challenge arises from the ND setup, where the absence of out-of-distribution (OOD) samples during training causes the detector to be biased toward the dominant style features in the in-distribution (ID) data. As a result, the model mistakenly learns to correlate style with core features, using this shortcut for detection. Robust ND is crucial for real-world applications like autonomous driving and medical imaging, where test samples may have different styles than the training data. Motivated by this, we propose a robust ND method that crafts an auxiliary OOD set with style features similar to the ID set but with different core features. Then, a task-based knowledge distillation strategy is utilized to distinguish core features from style features and help our model rely on core features for discriminating crafted OOD and ID sets. We verified the effectiveness of our method through extensive experimental evaluations on several datasets, including synthetic and real-world benchmarks, against nine different ND methods.
title A Contrastive Teacher-Student Framework for Novelty Detection under Style Shifts
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2501.17289