You are here: Artificial Intelligence > Deep Learning > Deep Learning Architectures

Deep Learning Architectures

The Model Generator dialog for the Deep Learning Tool and the Segmentation Wizard provides options for generating new Deep Learning models with a number of different architectures.

Click the New button on the Model Overview panel in the Deep Learning Tool or on the Models tab on the Segmentation Wizard panel to open the Model Generator dialog, shown below.

Model Generator dialog for the Deep Learning Tool

The following table lists the deep learning architectures available for semantic segmentation, super-resolution, and denoising. Note that the Segmentation Wizard only provides options for selecting architectures implemented for semantic segmentation.

Deep Learning architectures
	User for	Description
Attention U-Net	Semantic segmentation	This attention gate (AG) model, which was originally designed for medical imaging segmentation, automatically learns to focus on target structures of varying shapes and sizes while suppressing irrelevant regions in input images. By highlighting salient features only, the necessity of using explicit external tissue/organ localization module of cascaded convolutional neural networks (CNNs) is eliminated. Integrated into the standard U-Net architecture, AGs can increase model sensitivity and prediction accuracy. Reference… Oktay et al. Attention U-Net: Learning Where to Look for the Pancreas, arXiv, May 20, 2018 (also available online at https://arxiv.org/pdf/1804.03999.pdf).
Auto-Encoder	Semantic segmentation, denoising	Generic autoencoder. Reference… Additional information about Autoencoder is available online at: https://en.wikipedia.org/wiki/Autoencoder.
DeepLabV3+	Semantic segmentation	Reference… Chen et al. Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation, arXiv, August 22, 2018 (also available online at https://arxiv.org/pdf/1802.02611.pdf).
EDSR	Super-resolution	Reference… Lim et el. Enhanced Deep Residual Networks for Singe Image Super-Resolution, arXiv, July 10, 2017 (also available online at https://arxiv.org/pdf/1707.02921.pdf).
FC-DenseNet	Semantic segmentation	Reference… Jégou et al. The One Hundred Layers Tiramisu: Fully Convolutional DenseNets for Semantic Segmentation, arXiv, October 31, 2017 (also available online at https://arxiv.org/pdf/1611.09326.pdf).
LinkNet	Semantic segmentation	This architecture focuses on speed and efficiency for semantic segmentation tasks. Compared to other algorithms, LinkNet can learn with a more limited number of parameters and operations and still deliver accurate results. Reference… Abhishek Chaurasia and Eugenio Culurciello, LinkNet: Exploiting Encoder Representations for Efficient Semantic Segmentation, arXiv, June 14, 2017 (also available online at https://arxiv.org/pdf/1611.09326.pdf).
Noise2Noise	Denoising	Reference… Lehtinen et al. Noise2Noise: Learning Image Restoration without Clean Data, arXiv, October 29, 2018 (also available online at https://arxiv.org/pdf/1803.04189.pdf).
Noise2Noise_SRResNet	Denoising	Reference… Lehtinen et al. Noise2Noise: Learning Image Restoration without Clean Data, arXiv, October 29, 2018 (also available online at https://arxiv.org/pdf/1803.04189.pdf).
PSPNet	Semantic segmentation	Reference… Zhao et al. Pyramid Scene Parsing Network, arXiv, April 27, 2017 (also available online at: https://arxiv.org/pdf/1612.01105.pdf).
Sensor3D	Semantic segmentation	Semantic segmentation model using convolution LSTM. Reference… Novikov et al. Deep Sequential Segmentation of Organs in Volumetric Medical Scans, arXiv, March 11, 2019 (also available online at https://arxiv.org/pdf/1807.02437.pdf).
U-Net	Semantic segmentation, super-resolution, denoising	All purpose model designed especially for medical image segmentation. Reference… Ronneberger et al. U-Net: Convolutional Networks for Biomedical Image Segmentation, arXiv, May 18, 2015 (also available online at https://arxiv.org/pdf/1505.04597.pdf).
U-Net 3D	Semantic segmentation, super-resolution, denoising	3D implementation of U-Net. Reference… Ronneberger et al. U-Net: Convolutional Networks for Biomedical Image Segmentation, arXiv, May 18, 2015 (also available online at https://arxiv.org/pdf/1505.04597.pdf). Note Currently, only U-Net 3D is a fully 3D model that uses 3D convolutions. The number of input slices for this model is determined by the input size, which must be cubic. For example, 32x32x32. U-Net uses 2D convolutions, but can take 3D input patches for which you can choose the number of slices. You should also note that in some cases, 3D models can be more reliable for segmentation tasks.
U-Net++	Semantic segmentation, super-resolution, denoising	U-Net++ is a powerful architecture for medical image and semantic segmentation. This architecture is a deeply-supervised encoder-decoder network in which the encoder and decoder sub-networks are connected through a series of nested, dense skip pathways. The skip pathways help reduce the semantic gap between the feature maps of the encoder and decoder sub-networks. Reference… Zhou et al. UNet++: A Nested U-Net Architecture for Medical Image Segmentation, arXiv, July 18, 2018.
WDSR	Super-resolution	Reference… Yu et al. Wide Activation for Efficient and Accurate Image Super-Resolution, arXiv, December 21, 2018 (also available online at https://arxiv.org/pdf/1808.08718.pdf).

Editable Parameters for Deep Learning Architectures

A limited number of the basic parameters for Deep Learning architectures are available for editing in the Model Generator. These are listed below.

Refer to the documents referenced in the table above for more information about the editable parameters available for the implemented architectures.

You can also edit the basic and advanced training parameters of a Deep Learning model after it is generated (see Training Parameters for Deep Learning Models).

Editable parameters for Deep Learning model architectures
	Description
Attention U-Net	Depth level… Depth of the network, as determined by the number of pooling layers. Initial filter count… Filter count at the first convolution layer.
Autoencoder	Initial filter count… Filter count at the first convolutional layer. Kernel size… Convolutional filters kernel size. Pooling size… Pooling window size.
DeepLabV3+	Backbone… Backbone to use — Xception or MobileNetV2. Patch size… Fixed size of the input patches. Output stride… Ration of the image size to the encoder output size.
EDSR	Scale… Ration of the input size to the output size. Patch size… Fixed size of the input patches. Filter count… Filter count at each convolution layer. ResNet block count… The number of times to repeat ResNet blocks. Use Tanh activation… Determines if Tanh activation will be applied — True or False.
FC-DenseNet	Model type… Model variation to be generated — FC-DenseNet56, FC-DenseNet67, or FC-DenseNet103.
LinNet	Patch size… Fixed size of the input patches. Initial filter count… Filter count at the first convolution layer.
Noise2Noise	Initial filter count… Filter count at the first convolution layer.
Noise2Noise_SRResNet	Filter count… Filter count at each convolution layer. ResNet block count… The number of times to repeat ResNet blocks.
PSPNet	Backbone… Backbone to use — ResNet50 or ResNet101. Patch size… Fixed size of the input patches. Filter count… Filter count at each convolution layer.
Sensor3D	Depth level… Depth of the network, as determined by the number of pooling layers. Initial filter count… Filter count at the first convolution layer.
U-Net	Depth level… Depth of the network, as determined by the number of pooling layers. Initial filter count… Filter count at the first convolution layer.
U-Net 3D	Topology… The topology of the model. Initial filter count… Filter count at the first convolution layer. Use batch normalization… Determines if batch normalization will be applied — True or False.
U-Net	Depth level… Depth of the network, as determined by the number of pooling layers. Initial filter count… Filter count at the first convolution layer.
WDSR	Model type… Model variation to be generated — WDSR-A or WDSR-B. Scale… Ration of the input size to the output size. Patch size… Fixed size of the input patches. Filter count… Filter count at each convolution layer. ResNet block count… The number of times to repeat ResNet blocks. ResNet block expansion… The ratio to multiple the number of filters in the ResNet block expansion layer.