Author:
Alekhina Anna E.,Dorrer Mikhail G.,Ovchinnikov Alexander G.
Abstract
The article discusses the solution to the problem of choosing the architecture of a convolutional neural network for use in the computer vision of a smart vending refrigerator. Comparative tests decided the architectures of convolutional neural networks YOLOv2, YOLOv3, YOLOv4, Mask R-CNN, and YOLACT ++ on a standard MS COCO dataset, and then on datasets formed from images of typical smart refrigerator products. As a result of comparative tests, the best performance was demonstrated by the YOLOv3 architecture, trained based on a normalized dataset, supplemented with examples with complex intersections of samples without preprocessing examples. The obtained results substantiated the architecture used in computer vision of serially produced "smart" vending machines.
Reference13 articles.
1. Redmon J. and Farhadi A., You Only Look Once: Unified, Real-Time Object Detection 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2016)
2. Redmon J. and Farhadi A., YOLOv3: An Incremental Improvement arXiv:1804.02767 [cs.CV] (2018)
3. Bochkovskiy A., Wang C.-Y. and Liao H.-Y. M., YOLOv4: Optimal Speed and Accuracy of Object Detection arXiv:2004.10934 [cs.CV] (2020)
4. He K., Gkioxari G., Dollár P. and Girshick R., Mask R-CNN arXiv: 1703.06870 [cs.CV] (2017)
5. Bolya D., Zhou C., Xiao F. and Lee Y. J., YOLACT++ Better Real-Time Instance Segmentation IEEE Transactions on Pattern Analysis and Machine Intelligence ( Volume: 44, Issue: 2, 01 February 2022) (2019)