Weight-aware semi-supervised self-ensembling framework for interior decoration style classification.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Automatic interior decoration style classification is valuable but hindered by limited expert annotations.
• Existing models struggle with accuracy due to insufficient labeled data.

Purpose of the Study:

• To develop an accurate interior decoration style recognition model by effectively leveraging abundant unlabeled data.
• To address the challenge of data scarcity in supervised learning for design applications.

Main Methods:

• A weight-aware semi-supervised self-ensembling framework was developed.
• A weight module using a truncated Gaussian function adaptively assigns reliability scores to unlabeled data.
• Weighted consistency regularization, relation consistency, and class-aware contrastive learning were employed.

Main Results:

• The proposed framework significantly improves interior decoration style recognition performance.
• Experimental results demonstrate superior performance compared to existing semi-supervised learning methods.
• The method enhances model generalizability through synergistic regularization techniques.

Conclusions:

• The weight-aware semi-supervised self-ensembling framework effectively utilizes unlabeled data for style recognition.
• This approach offers a robust solution for accurate interior decoration style classification.
• The findings advance the field of automated design and computer vision applications.