Towards Unsupervised Domain Generalization

Zhang, Xingxuan; Zhou, Linjun; Xu, Renzhe; Cui, Peng; Shen, Zuo‐Jun Max; Liu, Haoxin

doi:10.1109/cvpr52688.2022.00486

Cited by 22 publications

(7 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Modern object detectors achieve striking performance when train and test data are sampled from the same or similar distributions (Zhang et al, 2022). However, if source and target domains differ, the generalization abilities of state-ofthe-art detectors lack (Y.…”

Section: A Object Detectionmentioning

confidence: 99%

“…However, if source and target domains differ, the generalization abilities of state-ofthe-art detectors lack (Y. Zhang et al, 2021Zhang et al, , 2022. One main reason described in the literature is the fact, that an object's background (= context) is related to the object itself (Zhang et al, 2022).…”

Section: A Object Detectionmentioning

confidence: 99%

“…Zhang et al, 2021Zhang et al, , 2022. One main reason described in the literature is the fact, that an object's background (= context) is related to the object itself (Zhang et al, 2022). Recent studies investigate domain generalization in object detection (DGOD), where detectors trained in source domains are evaluated on unknown target domains (Zhang et al, 2022).…”

Section: A Object Detectionmentioning

confidence: 99%

“…One main reason described in the literature is the fact, that an object's background (= context) is related to the object itself (Zhang et al, 2022). Recent studies investigate domain generalization in object detection (DGOD), where detectors trained in source domains are evaluated on unknown target domains (Zhang et al, 2022). To the best of our knowledge, there are no approaches that use partial data from one target domain class, to improve the generalization capabilities of knowledge previously learned from source domain data.…”

Section: A Object Detectionmentioning

confidence: 99%

“…Second, a (small) target domain dataset is used to adapt the model to the target domain. These approaches are particularly data efficient regarding the target domain since often a small target domain dataset is sufficient if the model can generalize well enough across the domains (Jaiswal et al, 2020;Pan & Yang, 2010;Shen et al, 2022;Zhang et al, 2022). However, like the conventional classifiers, these approaches rely on target domain data from all classes, so they can only be trained after the initial occurrence of visually detectable wear too (Figure 1).…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Cross-domain Transfer of Defect Features in Technical Domains based on Partial Target Data

Schlagenhauf¹,

Scheurenbrand²

2022

Preprint

View full text Add to dashboard Cite

A common challenge in real-world classification scenarios with sequentially appending target domain data is insufficient training datasets during the training phase. Therefore, conventional deep learning and transfer learning classifiers are not applicable especially when individual classes are not represented or are severely underrepresented at the outset. Domain Generalization approaches reach their limits when domain shifts become too large, making them occasionally unsuitable as well. In many (technical) domains, however, it is only the defect/ worn/ reject classes that are insufficiently represented, while the non-defect class is often available from the beginning. The proposed classification approach addresses such conditions and is based on a CNN encoder. Following a contrastive learning approach, it is trained with a modified triplet loss function using two datasets: Besides the non-defective target domain class (= 1st dataset), a state-of-the-art labeled source domain dataset that contains highly related classes (e.g., a related manufacturing error or wear defect) but originates from a (highly) different domain (e.g., different product, material, or appearance) (= 2nd dataset) is utilized. The approach learns the classification features from the source domain dataset while at the same time learning the differences between the source and the target domain in a single training step, aiming to transfer the relevant features to the target domain. The classifier becomes sensitive to the classification features and – by architecture – robust against the highly domain-specific context. The approach is benchmarked in a technical and a non-technical domain and shows convincing classification results. In particular, it is shown that the domain generalization capabilities and classification results are improved by the proposed architecture, allowing for larger domain shifts between source and target domains.

show abstract

Section: A Object Detectionmentioning

confidence: 99%