Condition-CNN: A hierarchical multi-label fashion image classification model
作者：
Highlights：
• Fashion image recognition involves prediction of a hierarchy of class labels.
• Branching Convolutional Neural Networks can predict hierarchical labels of an image.
• Condition-CNN is proposed based on Branching Convolutional Neural Networks.
• Kaggle Fashion Product Images dataset is used which has 3 levels of class labels.
• Achieves 99.8, 98.1, and 91.0% accuracies respectively for high to low level labels.
• Applies teacher forcing algorithm to reduce training time for greater accuracy.
摘要
•Fashion image recognition involves prediction of a hierarchy of class labels.•Branching Convolutional Neural Networks can predict hierarchical labels of an image.•Condition-CNN is proposed based on Branching Convolutional Neural Networks.•Kaggle Fashion Product Images dataset is used which has 3 levels of class labels.•Achieves 99.8, 98.1, and 91.0% accuracies respectively for high to low level labels.•Applies teacher forcing algorithm to reduce training time for greater accuracy.
论文关键词：Condition-CNN,Branching convolutional neural networks,Image classification,Convolutional neural networks,Hierarchical image classification,Teacher Forcing
论文评审过程：Received 6 November 2020, Revised 31 March 2021, Accepted 9 May 2021, Available online 12 May 2021, Version of Record 24 May 2021.
论文官网地址：https://doi.org/10.1016/j.eswa.2021.115195
