Open AccessClothing Image Detection and Recognition Based on Faster R-CNN
Author(s) -
Shuaifei Ji,
Runping Han,
Jianfeng Wei,
Rui Wang
Publication year - 2020
Publication title -
iop conference series. materials science and engineering
Language(s) - English
Resource type - Journals
eISSN - 1757-899X
pISSN - 1757-8981
DOI - 10.1088/1757-899x/790/1/012141
Subject(s) - softmax function , sigmoid function , computer science , convolutional neural network , bounding overwatch , pattern recognition (psychology) , artificial intelligence , image (mathematics) , feature (linguistics) , feature extraction , artificial neural network , activation function , linguistics , philosophy
Clothing image detection and recognition algorithm is proposed in this paper, which involves the design of two neural network models based on improved Faster R-CNN [1]. In our work, in order to train neural networks, ModaNet [2] dataset is selected and improvements on it are made, that is, the improper bounding boxes of footwear and boots in ModaNet dataset are modified based on their polygon annotations. Moreover, by making the following improvements on Faster R-CNN, our two neural network models are established, which use VGG16[3] and the first 40 layers of ResNet50 [4] as feature extration network, respectively. When using ResNet50 for feature extraction, the last 10 layers of ResNet50 are selected as part of the final classification and regression network. In addition, the softmax activation function in RPN is replaced with sigmoid. Through experimental comparison, it is found that both of these two models can well accomplish the task.