Details
Poster
Presenter(s)
![Yuma Yoshimoto Headshot](https://confcats-catavault.s3.amazonaws.com/CATAVault/ieeecass/master/files/styles/cc_user_photo/s3/user-pictures/22341_0.jpg?h=db5468e8&itok=YTUfQuLT)
Display Name
Yuma Yoshimoto
- Affiliation
-
AffiliationKyushu Institute of Technology
- Country
Abstract
The paper proposes Binarized Dual Stream VGG-16 (BDS-VGG16) model for service robots. Service robots require object recognition. The model is one of the convolutional neural network model for object recognition. It is hardware-oriented Dual Stream VGG-16 (DS-VGG16). It uses RGB and depth images. With the concept of Binarized Neural Networks, BDS-VGG16 is effective when implemented on FPGA. In results, the accuracy of BDS-VGG16 is 99.2%. In addition, we reduced the memory size required for calculating the coupling weight to 1/32 of the DS-VGG16 and reduced the required multiplications from 46 million to 0. It dramatically approached hardware implementation.