Figure 2 - available via license: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International
Content may be subject to copyright.
Download
View publication
Illustration of CIFAR-100 dataset examples. CIFAR-100 contains tiny images of 100 classes, with a resolution of 32x32. Shah et al. [28] managed to train CIFAR-100 using different ResNet models, including their variation, where ELU (Exponential Linear Unit) [29] was used as an activation function. Their test error on standard ResNet-101 achieved 27.23%. For this reason, we decided to use residual networks in our experiments.

Illustration of CIFAR-100 dataset examples. CIFAR-100 contains tiny images of 100 classes, with a resolution of 32x32. Shah et al. [28] managed to train CIFAR-100 using different ResNet models, including their variation, where ELU (Exponential Linear Unit) [29] was used as an activation function. Their test error on standard ResNet-101 achieved 27.23%. For this reason, we decided to use residual networks in our experiments.

Source publication
Figure 2: Illustration of CIFAR-100 dataset examples. CIFAR-100...
Construction of FatNet from Resnet-18
Comparison of the test accuracy and number of convolution operations...
FatNet: High Resolution Kernels for Classification Using Fully Convolutional Optical Neural Networks
Preprint
Full-text available
  • Oct 2022
This paper describes the transformation of a traditional in-silico classification network into an optical fully convolutional neural network with high-resolution feature maps and kernels. When using the free-space 4f system to accelerate the inference speed of neural networks, higher resolutions of feature maps and kernels can be used without the l...
Cite
Download full-text

Context in source publication

Context 1
... trained our network with the CIFAR-100 dataset (see Figure 2) and chose the ResNet-18 as the backbone network. The CIFAR-100 (Canadian Institute For Advanced Research) dataset consists of 60,000 images of 32x32 resolution. ...

Similar publications

Graphical representation of the 4f system performing the convolution...
Illustration of CIFAR-100 dataset examples. CIFAR-100 contains tiny...
Architecture comparison of our modified ResNet-18 used to train...
Training and validation accuracy for each experimented network at every...
+3 Graphical representation of the 4f system performing the convolution...
FatNet: High-Resolution Kernels for Classification Using Fully Convolutional Optical Neural Networks
Article
Full-text available
  • Apr 2023
This paper describes the transformation of a traditional in silico classification network into an optical fully convolutional neural network with high-resolution feature maps and kernels. When using the free-space 4f system to accelerate the inference speed of neural networks, higher resolutions of feature maps and kernels can be used without the l...
View