TY - GEN
T1 - Unsupervised Cross-domain Image Classification by Distance Metric Guided Feature Alignment
AU - Meng, Qingjie
AU - Rueckert, Daniel
AU - Kainz, Bernhard
PY - 2020/10/1
Y1 - 2020/10/1
N2 - Learning deep neural networks that are generalizable across different domains remains a challenge due to the problem of domain shift. Unsupervised domain adaptation is a promising avenue which transfers knowledge from a source domain to a target domain without using any labels in the target domain. Contemporary techniques focus on extracting domain-invariant features using domain adversarial training. However, these techniques neglect to learn discriminative class boundaries in the latent representation space on a target domain and yield limited adaptation performance. To address this problem, we propose distance metric guided feature alignment (MetFA) to extract discriminative as well as domain-invariant features on both source and target domains. The proposed MetFA method explicitly and directly learns the latent representation without using domain adversarial training. Our model integrates class distribution alignment to transfer semantic knowledge from a source domain to a target domain. We evaluate the proposed method on fetal ultrasound datasets for cross-device image classification. Experimental results demonstrate that the proposed method outperforms the state-of-the-art and enables model generalization.
AB - Learning deep neural networks that are generalizable across different domains remains a challenge due to the problem of domain shift. Unsupervised domain adaptation is a promising avenue which transfers knowledge from a source domain to a target domain without using any labels in the target domain. Contemporary techniques focus on extracting domain-invariant features using domain adversarial training. However, these techniques neglect to learn discriminative class boundaries in the latent representation space on a target domain and yield limited adaptation performance. To address this problem, we propose distance metric guided feature alignment (MetFA) to extract discriminative as well as domain-invariant features on both source and target domains. The proposed MetFA method explicitly and directly learns the latent representation without using domain adversarial training. Our model integrates class distribution alignment to transfer semantic knowledge from a source domain to a target domain. We evaluate the proposed method on fetal ultrasound datasets for cross-device image classification. Experimental results demonstrate that the proposed method outperforms the state-of-the-art and enables model generalization.
UR - https://www.scopus.com/pages/publications/85092739330
U2 - 10.1007/978-3-030-60334-2_15
DO - 10.1007/978-3-030-60334-2_15
M3 - Conference contribution
SN - 9783030603335
T3 - Lecture Notes in Computer Science
SP - 146
EP - 157
BT - Medical Ultrasound, and Preterm, Perinatal and Paediatric Image Analysis
PB - Springer
T2 - First International Workshop, ASMUS 2020, and 5th International Workshop, PIPPI 2020
Y2 - 4 October 2020 through 8 October 2020
ER -