Part-based annotation-free fine-grained classification of images of retail products

作者：

Highlights：

• A deep convolutional reconstruction-classification network,RC-Net is introduced for fine-grained classification of products.

• An annotation-free conv-LSTM based part-level classifier is proposed to classify discriminative parts of the products.

• The discriminative parts of the products are identified in a unique unsupervised technique.

• Proposed fine-grained classifier significantly improves the performance of R-CNN for detecting retail products.

摘要

•A deep convolutional reconstruction-classification network,RC-Net is introduced for fine-grained classification of products.•An annotation-free conv-LSTM based part-level classifier is proposed to classify discriminative parts of the products.•The discriminative parts of the products are identified in a unique unsupervised technique.•Proposed fine-grained classifier significantly improves the performance of R-CNN for detecting retail products.

论文关键词：Fine-grained classification,Reconstruction-classification network,Supervised convolutional autoencoder,Retail product detection

论文评审过程：Received 26 January 2021, Revised 30 July 2021, Accepted 13 August 2021, Available online 14 August 2021, Version of Record 26 August 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.108257