Knowledge distillation methods for efficient unsupervised adaptation across multiple domains

Nguyen-Meidine, Le Thanh; Belal, Atif; Kiran, Madhu; Dolz, José; Blais-Morin, Louis-Antoine; Granger, Éric

doi:10.1016/j.imavis.2021.104096

“…These methods seek to adapt CNNs trained with annotated source video data to perform well in a target domain by leveraging unlabeled data captured from that domain. To learn a discriminant domain-invariant feature representation from source and target data, STDA methods typically rely on discrepancy-based or adversarial approaches [3,5,6,7,8,9,10,11,12,13].…”

Section: Introductionmentioning

confidence: 99%

“…Nevertheless, these approaches are either too complex (require one model per target domain), or generalize poorly on distinct target domains, particularly when adapting a smaller common CNN backbone on a growing number of targets. In [6], MTDA is performed by distilling information from targetspecific teachers to a student model (deployed for testing), significantly reducing system complexity.…”

Section: Introductionmentioning

confidence: 99%

Knowledge Distillation for Multi-Target Domain Adaptation in Real-Time Person Re-Identification

Remigereau

¹

,

Mekhazni

²

,

Abdoli

³

et al. 2022

2022 IEEE International Conference on Image Processing (ICIP)

View full text Add to dashboard Cite

Despite the recent success of deep learning architectures, person re-identification (ReID) remains a challenging problem in real-word applications. Several unsupervised single-target domain adaptation (STDA) methods have recently been proposed to limit the decline in ReID accuracy caused by the domain shift that typically occurs between source and target video data. Given the multimodal nature of person ReID data (due to variations across camera viewpoints and capture conditions), training a common CNN backbone to address domain shifts across multiple target domains, can provide an efficient solution for real-time ReID applications. Although multi-target domain adaptation (MTDA) has not been widely addressed in the ReID literature, a straightforward approach consists in blending different target datasets, and performing STDA on the mixture to train a common CNN. However, this approach may lead to poor generalization, especially when blending a growing number of distinct target domains to train a smaller CNN. To alleviate this problem, we introduce a new MTDA method based on knowledge distillation (KD-ReID) that is suitable for real-time person ReID applications. Our method adapts a common lightweight student backbone CNN over the target domains by alternatively distilling from multiple specialized teacher CNNs, each one adapted on data from a specific target domain. Extensive experiments 1 conducted on several challenging person ReID datasets indicate that our approach outperforms state-of-art methods for MTDA, including blending methods, particularly when training a compact CNN backbone like OSNet. Results suggest that our flexible MTDA approach can be employed to design cost-effective ReID systems for real-time video surveillance applications.

show abstract

“…These methods seek to adapt CNNs trained with annotated source video data to perform well in a target domain by leveraging unlabeled data captured from that domain. To learn a discriminant domaininvariant feature representation from source and target data, STDA methods rely on, e.g., discrepancy-based or adversarial approaches [5,3,6,7,8,9,10,11,12,13].…”

Section: Introductionmentioning

confidence: 99%

“…Nevertheless, these approaches are either too complex it requires one model per target domain, or generalize poorly on distinct target domains, particularly when adapting a smaller common CNN backbone on a growing number of targets. In [6], MTDA is performed by distilling information from targetspecific teachers to a student model (deployed for testing), significantly reducing system complexity.…”

Section: Introductionmentioning

confidence: 99%

Knowledge Distillation for Multi-Target Domain Adaptation in Real-Time Person Re-Identification

Remigereau¹,

Mekhazni²,

Abdoli³

et al. 2022

Preprint

Self Cite

0

View full text Add to dashboard Cite

Despite the recent success of deep learning architectures, person re-identification (ReID) remains a challenging problem in real-word applications. Several unsupervised single-target domain adaptation (STDA) methods have recently been proposed to limit the decline in ReID accuracy caused by the domain shift that typically occurs between source and target video data. Given the multimodal nature of person ReID data (due to variations across camera viewpoints and capture conditions), training a common CNN backbone to address domain shifts across multiple target domains, can provide an efficient solution for real-time ReID applications. Although multi-target domain adaptation (MTDA) has not been widely addressed in the ReID literature, a straightforward approach consists in blending different target datasets, and performing STDA on the mixture to train a common CNN. However, this approach may lead to poor generalization, especially when blending a growing number of distinct target domains to train a smaller CNN. To alleviate this problem, we introduce a new MTDA method based on knowledge distillation (KD-ReID) that is suitable for real-time person ReID applications. Our method adapts a common lightweight student backbone CNN over the target domains by alternatively distilling from multiple specialized teacher CNNs, each one adapted on data from a specific target domain. Extensive experiments 1 conducted on several challenging person ReID datasets indicate that our approach outperforms state-of-art methods for MTDA, including blending methods, particularly when training a compact CNN backbone like OSNet. Results suggest that our flexible MTDA approach can be employed to design cost-effective ReID systems for real-time video surveillance applications.

show abstract

Continual Source-Free Unsupervised Domain Adaptation

Ahmed,

Morerio,

Murino

2023

Lecture Notes in Computer Science

3

0

View full text Add to dashboard Cite

Knowledge distillation methods for efficient unsupervised adaptation across multiple domains

Cited by 14 publications

References 5 publications

Knowledge Distillation for Multi-Target Domain Adaptation in Real-Time Person Re-Identification

Knowledge Distillation for Multi-Target Domain Adaptation in Real-Time Person Re-Identification

Knowledge Distillation for Multi-Target Domain Adaptation in Real-Time Person Re-Identification

Continual Source-Free Unsupervised Domain Adaptation

Contact Info

Product

Resources

About