- Media data analysis
Unsupervised image clustering: IDFD
Realizes strong performance on inspection image clustering at manufacturing site, and reduces costs of visual checking and manual data analysis.
- Accuracy for unsupervised clustering of public image data sets increased from 62.3% to 81.5% (the state-of-the-art perfirmance).
- Achieves highly accurate unsupervised clustering by learning feature preferable for clustering from complex images.
Applications
- Inspection images in semiconductor manufacturing
- Visual Inspection images in the manufacturing field
Benchmarks, strengths, and track record
- Achieved unsupervised clustering accuracy of 81.5% and 95.4% respectively for CIFAR-10 and ImageNet-10 image data sets. (62.3% and 71.0% using conventional methods)
- Achieved state-of-the-art unsupervised image clustering accuracy; adopted at ICLR2021.
Inquiries
Please include the title “Toshiba AI Technology Catalog: “Unsupervised image clustering technology IDFD” or the URL in the inquiry text.
Please note that because this technology is currently the subject of R&D activities, immediate responses to inquiries may not be possible.
References:
- Development of unsupervised image clustering AI, which learns clustering-friendly features from complex images (press release; April 28, 2021) (in Japanese)
- Y. Tao, K. Takagi, K. Nakata, “Clustering-friendly Representation Learning via Instance Discrimination and Feature Decorrelation,” The Ninth International Conference on Learning Representations (ICLR2021), 2021.
- Oshima, T., Takagi, K., Nakata, “Clustering-Friendly Representation Learning for Enhancing Salient Features”, 28th Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD2024), 2024.
