**Inception v3** is a convolutional neural network architecture from the Inception family that makes several improvements including using [Label Smoothing](https://paperswithcode.com/method/label-smoothing), Factorized 7 x 7 convolutions, and the use of an [auxiliary classifer](https://paperswithcode.com/method/auxiliary-classifier) to propagate label information lower down the network (along with the use of batch normalization for layers in the sidehead). The key building block is an [Inception Module](https://paperswithcode.com/method/inception-v3-module).
This particular model was trained for study of adversarial examples (adversarial training).
The weights from this model were ported from [Tensorflow/Models](https://github.com/tensorflow/models).
**AdvProp** is an adversarial training scheme which treats adversarial examples as additional examples, to prevent overfitting. Key to the method is the usage of a separate auxiliary batch norm for adversarial examples, as they have different underlying distributions to normal examples.
**Big Transfer (BiT)** is a type of pretraining recipe that pre-trains on a large supervised source dataset, and fine-tunes the weights on the target task. Models are trained on the JFT-300M dataset. The finetuned models contained in this collection are finetuned on ImageNet.
**CSPDarknet53** is a convolutional neural network and backbone for object detection that uses [DarkNet-53](https://paperswithcode.com/method/darknet-53). It employs a CSPNet strategy to partition the feature map of the base layer into two parts and then merges them through a cross-stage hierarchy. The use of a split and merge strategy allows for more gradient flow through the network.
**CSPResNet** is a convolutional neural network where we apply the Cross Stage Partial Network (CSPNet) approach to [ResNet](https://paperswithcode.com/method/resnet). The CSPNet partitions the feature map of the base layer into two parts and then merges them through a cross-stage hierarchy. The use of a split and merge strategy allows for more gradient flow through the network.
**CSPResNeXt** is a convolutional neural network where we apply the Cross Stage Partial Network (CSPNet) approach to [ResNeXt](https://paperswithcode.com/method/resnext). The CSPNet partitions the feature map of the base layer into two parts and then merges them through a cross-stage hierarchy. The use of a split and merge strategy allows for more gradient flow through the network.
**DenseNet** is a type of convolutional neural network that utilises dense connections between layers, through [Dense Blocks](http://www.paperswithcode.com/method/dense-block), where we connect *all layers* (with matching feature-map sizes) directly with each other. To preserve the feed-forward nature, each layer obtains additional inputs from all preceding layers and passes on its own feature-maps to all subsequent layers.
Extending “shallow” skip connections, **Dense Layer Aggregation (DLA)** incorporates more depth and sharing. The authors introduce two structures for deep layer aggregation (DLA): iterative deep aggregation (IDA) and hierarchical deep aggregation (HDA). These structures are expressed through an architectural framework, independent of the choice of backbone, for compatibility with current and future networks.
A **Dual Path Network (DPN)** is a convolutional neural network which presents a new topology of connection paths internally. The intuition is that [ResNets](https://paperswithcode.com/method/resnet) enables feature re-usage while DenseNet enables new feature exploration, and both are important for learning good representations. To enjoy the benefits from both path topologies, Dual Path Networks share common features while maintaining the flexibility to explore new features through dual path architectures.
An **ECA ResNet** is a variant on a [ResNet](https://paperswithcode.com/method/resnet) that utilises an [Efficient Channel Attention module](https://paperswithcode.com/method/efficient-channel-attention). Efficient Channel Attention is an architectural unit based on [squeeze-and-excitation blocks](https://paperswithcode.com/method/squeeze-and-excitation-block) that reduces model complexity without dimensionality reduction.
**EfficientNet** is a convolutional neural network architecture and scaling method that uniformly scales all dimensions of depth/width/resolution using a *compound coefficient*. Unlike conventional practice that arbitrary scales these factors, the EfficientNet scaling method uniformly scales network width, depth, and resolution with a set of fixed scaling coefficients. For example, if we want to use $2^N$ times more computational resources, then we can simply increase the network depth by $\alpha ^ N$, width by $\beta ^ N$, and image size by $\gamma ^ N$, where $\alpha, \beta, \gamma$ are constant coefficients determined by a small grid search on the original small model. EfficientNet uses a compound coefficient $\phi$ to uniformly scales network width, depth, and resolution in a principled way.
@ -89,14 +89,13 @@ You can follow the [timm recipe scripts](https://rwightman.github.io/pytorch-ima
**EfficientNet** is a convolutional neural network architecture and scaling method that uniformly scales all dimensions of depth/width/resolution using a *compound coefficient*. Unlike conventional practice that arbitrary scales these factors, the EfficientNet scaling method uniformly scales network width, depth, and resolution with a set of fixed scaling coefficients. For example, if we want to use $2^N$ times more computational resources, then we can simply increase the network depth by $\alpha ^ N$, width by $\beta ^ N$, and image size by $\gamma ^ N$, where $\alpha, \beta, \gamma$ are constant coefficients determined by a small grid search on the original small model. EfficientNet uses a compound coefficient $\phi$ to uniformly scales network width, depth, and resolution in a principled way.
**Inception-ResNet-v2** is a convolutional neural architecture that builds on the Inception family of architectures but incorporates [residual connections](https://paperswithcode.com/method/residual-connection) (replacing the filter concatenation stage of the Inception architecture).
This particular model was trained for study of adversarial examples (adversarial training).
The weights from this model were ported from [Tensorflow/Models](https://github.com/tensorflow/models).
**VoVNet** is a convolutional neural network that seeks to make [DenseNet](https://paperswithcode.com/method/densenet) more efficient by concatenating all features only once in the last feature map, which makes input size constant and enables enlarging new output channel.
**FBNet** is a type of convolutional neural architectures discovered through [DNAS](https://paperswithcode.com/method/dnas) neural architecture search. It utilises a basic type of image model block inspired by [MobileNetv2](https://paperswithcode.com/method/mobilenetv2) that utilises depthwise convolutions and an inverted residual structure (see components).
**Inception v3** is a convolutional neural network architecture from the Inception family that makes several improvements including using [Label Smoothing](https://paperswithcode.com/method/label-smoothing), Factorized 7 x 7 convolutions, and the use of an [auxiliary classifer](https://paperswithcode.com/method/auxiliary-classifier) to propagate label information lower down the network (along with the use of batch normalization for layers in the sidehead). The key building block is an [Inception Module](https://paperswithcode.com/method/inception-v3-module).
The weights from this model were ported from Gluon.
The weights from this model were ported from [Gluon](https://cv.gluon.ai/model_zoo/classification.html).
**Residual Networks**, or **ResNets**, learn residual functions with reference to the layer inputs, instead of learning unreferenced functions. Instead of hoping each few stacked layers directly fit a desired underlying mapping, residual nets let these layers fit a residual mapping. They stack [residual blocks](https://paperswithcode.com/method/residual-block) ontop of each other to form network: e.g. a ResNet-50 has fifty layers using these blocks.
The weights from this model were ported from Gluon.
The weights from this model were ported from [Gluon](https://cv.gluon.ai/model_zoo/classification.html).
A **ResNeXt** repeats a [building block](https://paperswithcode.com/method/resnext-block) that aggregates a set of transformations with the same topology. Compared to a [ResNet](https://paperswithcode.com/method/resnet), it exposes a new dimension, *cardinality* (the size of the set of transformations) $C$, as an essential factor in addition to the dimensions of depth and width.
The weights from this model were ported from Gluon.
The weights from this model were ported from [Gluon](https://cv.gluon.ai/model_zoo/classification.html).
A **SENet** is a convolutional neural network architecture that employs [squeeze-and-excitation blocks](https://paperswithcode.com/method/squeeze-and-excitation-block) to enable the network to perform dynamic channel-wise feature recalibration.
The weights from this model were ported from Gluon.
The weights from this model were ported from [Gluon](https://cv.gluon.ai/model_zoo/classification.html).
**SE ResNeXt** is a variant of a [ResNext](https://www.paperswithcode.com/method/resnext) that employs [squeeze-and-excitation blocks](https://paperswithcode.com/method/squeeze-and-excitation-block) to enable the network to perform dynamic channel-wise feature recalibration.
The weights from this model were ported from Gluon.
The weights from this model were ported from [Gluon](https://cv.gluon.ai/model_zoo/classification.html).
**Xception** is a convolutional neural network architecture that relies solely on [depthwise separable convolution](https://paperswithcode.com/method/depthwise-separable-convolution) layers. The weights from this model were ported from Gluon.
**Xception** is a convolutional neural network architecture that relies solely on [depthwise separable convolution](https://paperswithcode.com/method/depthwise-separable-convolution) layers.
The weights from this model were ported from [Gluon](https://cv.gluon.ai/model_zoo/classification.html).
**HRNet**, or **High-Resolution Net**, is a general purpose convolutional neural network for tasks like semantic segmentation, object detection and image classification. It is able to maintain high resolution representations through the whole process. We start from a high-resolution convolution stream, gradually add high-to-low resolution convolution streams one by one, and connect the multi-resolution streams in parallel. The resulting network consists of several ($4$ in the paper) stages and the $n$th stage contains $n$ streams corresponding to $n$ resolutions. The authors conduct repeated multi-resolution fusions by exchanging the information across the parallel streams over and over.
A **ResNeXt** repeats a [building block](https://paperswithcode.com/method/resnext-block) that aggregates a set of transformations with the same topology. Compared to a [ResNet](https://paperswithcode.com/method/resnet), it exposes a new dimension, *cardinality* (the size of the set of transformations) $C$, as an essential factor in addition to the dimensions of depth and width.
**Inception-ResNet-v2** is a convolutional neural architecture that builds on the Inception family of architectures but incorporates [residual connections](https://paperswithcode.com/method/residual-connection) (replacing the filter concatenation stage of the Inception architecture).
**Inception v3** is a convolutional neural network architecture from the Inception family that makes several improvements including using [Label Smoothing](https://paperswithcode.com/method/label-smoothing), Factorized 7 x 7 convolutions, and the use of an [auxiliary classifer](https://paperswithcode.com/method/auxiliary-classifier) to propagate label information lower down the network (along with the use of batch normalization for layers in the sidehead). The key building block is an [Inception Module](https://paperswithcode.com/method/inception-v3-module).
**Inception-v4** is a convolutional neural network architecture that builds on previous iterations of the Inception family by simplifying the architecture and using more inception modules than [Inception-v3](https://paperswithcode.com/method/inception-v3).
**SE ResNet** is a variant of a [ResNet](https://www.paperswithcode.com/method/resnet) that employs [squeeze-and-excitation blocks](https://paperswithcode.com/method/squeeze-and-excitation-block) to enable the network to perform dynamic channel-wise feature recalibration.
**SE ResNeXt** is a variant of a [ResNeXt](https://www.paperswithcode.com/method/resnext) that employs [squeeze-and-excitation blocks](https://paperswithcode.com/method/squeeze-and-excitation-block) to enable the network to perform dynamic channel-wise feature recalibration.
A **SENet** is a convolutional neural network architecture that employs [squeeze-and-excitation blocks](https://paperswithcode.com/method/squeeze-and-excitation-block) to enable the network to perform dynamic channel-wise feature recalibration.
**MixNet** is a type of convolutional neural network discovered via AutoML that utilises [MixConvs](https://paperswithcode.com/method/mixconv) instead of regular [depthwise convolutions](https://paperswithcode.com/method/depthwise-convolution).
**MnasNet** is a type of convolutional neural network optimized for mobile devices that is discovered through mobile neural architecture search, which explicitly incorporates model latency into the main objective so that the search can identify a model that achieves a good trade-off between accuracy and latency. The main building block is an [inverted residual block](https://paperswithcode.com/method/inverted-residual-block) (from [MobileNetV2](https://paperswithcode.com/method/mobilenetv2)).
**MobileNetV2** is a convolutional neural network architecture that seeks to perform well on mobile devices. It is based on an [inverted residual structure](https://paperswithcode.com/method/inverted-residual-block) where the residual connections are between the bottleneck layers. The intermediate expansion layer uses lightweight depthwise convolutions to filter features as a source of non-linearity. As a whole, the architecture of MobileNetV2 contains the initial fully convolution layer with 32 filters, followed by 19 residual bottleneck layers.
**MobileNetV3** is a convolutional neural network that is designed for mobile phone CPUs. The network design includes the use of a [hard swish activation](https://paperswithcode.com/method/hard-swish) and [squeeze-and-excitation](https://paperswithcode.com/method/squeeze-and-excitation-block) modules in the [MBConv blocks](https://paperswithcode.com/method/inverted-residual-block).
**NASNet** is a type of convolutional neural network discovered through neural architecture search. The building blocks consist of normal and reduction cells.
**Progressive Neural Architecture Search**, or **PNAS**, is a method for learning the structure of convolutional neural networks (CNNs). It uses a sequential model-based optimization (SMBO) strategy, where we search the space of cell structures, starting with simple (shallow) models and progressing to complex ones, pruning out unpromising structures as we go.
**RegNetX** is a convolutional network design space with simple, regular models with parameters: depth $d$, initial width $w\_{0} > 0$, and slope $w\_{a} > 0$, and generates a different block width $u\_{j}$ for each block $j <d$.ThekeyrestrictionfortheRegNettypesofmodelisthatthereisalinearparameterisationofblockwidths(thedesignspaceonlycontainsmodelswiththislinearstructure):
**RegNetY** is a convolutional network design space with simple, regular models with parameters: depth $d$, initial width $w\_{0} > 0$, and slope $w\_{a} > 0$, and generates a different block width $u\_{j}$ for each block $j <d$.ThekeyrestrictionfortheRegNettypesofmodelisthatthereisalinearparameterisationofblockwidths(thedesignspaceonlycontainsmodelswiththislinearstructure):
**Res2Net** is an image model that employs a variation on bottleneck residual blocks, [Res2Net Blocks](https://paperswithcode.com/method/res2net-block). The motivation is to be able to represent features at multiple scales. This is achieved through a novel building block for CNNs that constructs hierarchical residual-like connections within one single residual block. This represents multi-scale features at a granular level and increases the range of receptive fields for each network layer.
**Res2Net** is an image model that employs a variation on [ResNeXt](https://paperswithcode.com/method/resnext) bottleneck residual blocks. The motivation is to be able to represent features at multiple scales. This is achieved through a novel building block for CNNs that constructs hierarchical residual-like connections within one single residual block. This represents multi-scale features at a granular level and increases the range of receptive fields for each network layer.
**Res2NeXt** is an image model that employs a variation on [ResNeXt](https://paperswithcode.com/method/resnext) bottleneck residual blocks. The motivation is to be able to represent features at multiple scales. This is achieved through a novel building block for CNNs that constructs hierarchical residual-like connections within one single residual block. This represents multi-scale features at a granular level and increases the range of receptive fields for each network layer.
A **ResNest** is a variant on a [ResNet](https://paperswithcode.com/method/resnet), which instead stacks [Split-Attention blocks](https://paperswithcode.com/method/split-attention). The cardinal group representations are then concatenated along the channel dimension: $V = \text{Concat}${$V^{1},V^{2},\cdots{V}^{K}$}. As in standard residual blocks, the final output $Y$ of otheur Split-Attention block is produced using a shortcut connection: $Y=V+X$, if the input and output feature-map share the same shape. For blocks with a stride, an appropriate transformation $\mathcal{T}$ is applied to the shortcut connection to align the output shapes: $Y=V+\mathcal{T}(X)$. For example, $\mathcal{T}$ can be strided convolution or combined convolution-with-pooling.
A **ResNeSt** is a variant on a [ResNet](https://paperswithcode.com/method/resnet), which instead stacks [Split-Attention blocks](https://paperswithcode.com/method/split-attention). The cardinal group representations are then concatenated along the channel dimension: $V = \text{Concat}${$V^{1},V^{2},\cdots{V}^{K}$}. As in standard residual blocks, the final output $Y$ of otheur Split-Attention block is produced using a shortcut connection: $Y=V+X$, if the input and output feature-map share the same shape. For blocks with a stride, an appropriate transformation $\mathcal{T}$ is applied to the shortcut connection to align the output shapes: $Y=V+\mathcal{T}(X)$. For example, $\mathcal{T}$ can be strided convolution or combined convolution-with-pooling.
**ResNet-D** is a modification on the [ResNet](https://paperswithcode.com/method/resnet) architecture that utilises an [average pooling](https://paperswithcode.com/method/average-pooling) tweak for downsampling. The motivation is that in the unmodified ResNet, the [1×1 convolution](https://paperswithcode.com/method/1x1-convolution) for the downsampling block ignores 3/4 of input feature maps, so this is modified so no information will be ignored
**Residual Networks**, or **ResNets**, learn residual functions with reference to the layer inputs, instead of learning unreferenced functions. Instead of hoping each few stacked layers directly fit a desired underlying mapping, residual nets let these layers fit a residual mapping. They stack [residual blocks](https://paperswithcode.com/method/residual-block) ontop of each other to form network: e.g. a ResNet-50 has fifty layers using these blocks.
A **ResNeXt** repeats a [building block](https://paperswithcode.com/method/resnext-block) that aggregates a set of transformations with the same topology. Compared to a [ResNet](https://paperswithcode.com/method/resnet), it exposes a new dimension, *cardinality* (the size of the set of transformations) $C$, as an essential factor in addition to the dimensions of depth and width.
**Rank Expansion Networks** (ReXNets) follow a set of new design principles for designing bottlenecks in image classification models. Authors refine each layer by 1) expanding the input channel size of the convolution layer and 2) replacing the [ReLU6s](https://www.paperswithcode.com/method/relu6).
**SE ResNet** is a variant of a [ResNet](https://www.paperswithcode.com/method/resnet) that employs [squeeze-and-excitation blocks](https://paperswithcode.com/method/squeeze-and-excitation-block) to enable the network to perform dynamic channel-wise feature recalibration.
**SelecSLS** uses novel selective long and short range skip connections to improve the information flow allowing for a drastically faster network without compromising accuracy.
**SE ResNeXt** is a variant of a [ResNext](https://www.paperswithcode.com/method/resneXt) that employs [squeeze-and-excitation blocks](https://paperswithcode.com/method/squeeze-and-excitation-block) to enable the network to perform dynamic channel-wise feature recalibration.
**SK ResNet** is a variant of a [ResNet](https://www.paperswithcode.com/method/resnet) that employs a [Selective Kernel](https://paperswithcode.com/method/selective-kernel) unit. In general, all the large kernel convolutions in the original bottleneck blocks in ResNet are replaced by the proposed [SK convolutions](https://paperswithcode.com/method/selective-kernel-convolution), enabling the network to choose appropriate receptive field sizes in an adaptive manner.
**SK ResNeXt** is a variant of a [ResNeXt](https://www.paperswithcode.com/method/resnext) that employs a [Selective Kernel](https://paperswithcode.com/method/selective-kernel) unit. In general, all the large kernel convolutions in the original bottleneck blocks in ResNext are replaced by the proposed [SK convolutions](https://paperswithcode.com/method/selective-kernel-convolution), enabling the network to choose appropriate receptive field sizes in an adaptive manner.
**Residual Networks**, or **ResNets**, learn residual functions with reference to the layer inputs, instead of learning unreferenced functions. Instead of hoping each few stacked layers directly fit a desired underlying mapping, residual nets let these layers fit a residual mapping. They stack [residual blocks](https://paperswithcode.com/method/residual-block) ontop of each other to form network: e.g. a ResNet-50 has fifty layers using these blocks.
A **ResNeXt** repeats a [building block](https://paperswithcode.com/method/resnext-block) that aggregates a set of transformations with the same topology. Compared to a [ResNet](https://paperswithcode.com/method/resnet), it exposes a new dimension, *cardinality* (the size of the set of transformations) $C$, as an essential factor in addition to the dimensions of depth and width.
**Residual Networks**, or **ResNets**, learn residual functions with reference to the layer inputs, instead of learning unreferenced functions. Instead of hoping each few stacked layers directly fit a desired underlying mapping, residual nets let these layers fit a residual mapping. They stack [residual blocks](https://paperswithcode.com/method/residual-block) ontop of each other to form network: e.g. a ResNet-50 has fifty layers using these blocks.
A **ResNeXt** repeats a [building block](https://paperswithcode.com/method/resnext-block) that aggregates a set of transformations with the same topology. Compared to a [ResNet](https://paperswithcode.com/method/resnet), it exposes a new dimension, *cardinality* (the size of the set of transformations) $C$, as an essential factor in addition to the dimensions of depth and width.
**EfficientNet** is a convolutional neural network architecture and scaling method that uniformly scales all dimensions of depth/width/resolution using a *compound coefficient*. Unlike conventional practice that arbitrary scales these factors, the EfficientNet scaling method uniformly scales network width, depth, and resolution with a set of fixed scaling coefficients. For example, if we want to use $2^N$ times more computational resources, then we can simply increase the network depth by $\alpha ^ N$, width by $\beta ^ N$, and image size by $\gamma ^ N$, where $\alpha, \beta, \gamma$ are constant coefficients determined by a small grid search on the original small model. EfficientNet uses a compound coefficient $\phi$ to uniformly scales network width, depth, and resolution in a principled way.
@ -8,6 +8,8 @@ The base EfficientNet-B0 network is based on the inverted bottleneck residual bl
This collection of models amends EfficientNet by adding [CondConv](https://paperswithcode.com/method/condconv) convolutions.
The weights from this model were ported from [Tensorflow/TPU](https://github.com/tensorflow/tpu).
**EfficientNet** is a convolutional neural network architecture and scaling method that uniformly scales all dimensions of depth/width/resolution using a *compound coefficient*. Unlike conventional practice that arbitrary scales these factors, the EfficientNet scaling method uniformly scales network width, depth, and resolution with a set of fixed scaling coefficients. For example, if we want to use $2^N$ times more computational resources, then we can simply increase the network depth by $\alpha ^ N$, width by $\beta ^ N$, and image size by $\gamma ^ N$, where $\alpha, \beta, \gamma$ are constant coefficients determined by a small grid search on the original small model. EfficientNet uses a compound coefficient $\phi$ to uniformly scales network width, depth, and resolution in a principled way.
@ -8,6 +8,8 @@ The base EfficientNet-B0 network is based on the inverted bottleneck residual bl
EfficientNet-Lite makes EfficientNet more suitable for mobile devices by introducing [ReLU6](https://paperswithcode.com/method/relu6) activation functions and removing [squeeze-and-excitation blocks](https://paperswithcode.com/method/squeeze-and-excitation).
The weights from this model were ported from [Tensorflow/TPU](https://github.com/tensorflow/tpu).
**EfficientNet** is a convolutional neural network architecture and scaling method that uniformly scales all dimensions of depth/width/resolution using a *compound coefficient*. Unlike conventional practice that arbitrary scales these factors, the EfficientNet scaling method uniformly scales network width, depth, and resolution with a set of fixed scaling coefficients. For example, if we want to use $2^N$ times more computational resources, then we can simply increase the network depth by $\alpha ^ N$, width by $\beta ^ N$, and image size by $\gamma ^ N$, where $\alpha, \beta, \gamma$ are constant coefficients determined by a small grid search on the original small model. EfficientNet uses a compound coefficient $\phi$ to uniformly scales network width, depth, and resolution in a principled way.
@ -6,6 +6,8 @@ The compound scaling method is justified by the intuition that if the input imag
The base EfficientNet-B0 network is based on the inverted bottleneck residual blocks of [MobileNetV2](https://paperswithcode.com/method/mobilenetv2), in addition to [squeeze-and-excitation blocks](https://paperswithcode.com/method/squeeze-and-excitation-block).
The weights from this model were ported from [Tensorflow/TPU](https://github.com/tensorflow/tpu).
**Inception v3** is a convolutional neural network architecture from the Inception family that makes several improvements including using [Label Smoothing](https://paperswithcode.com/method/label-smoothing), Factorized 7 x 7 convolutions, and the use of an [auxiliary classifer](https://paperswithcode.com/method/auxiliary-classifier) to propagate label information lower down the network (along with the use of batch normalization for layers in the sidehead). The key building block is an [Inception Module](https://paperswithcode.com/method/inception-v3-module).
The weights from this model were ported from [Tensorflow/Models](https://github.com/tensorflow/models).
**MixNet** is a type of convolutional neural network discovered via AutoML that utilises [MixConvs](https://paperswithcode.com/method/mixconv) instead of regular [depthwise convolutions](https://paperswithcode.com/method/depthwise-convolution).
The weights from this model were ported from [Tensorflow/TPU](https://github.com/tensorflow/tpu).
**MobileNetV3** is a convolutional neural network that is designed for mobile phone CPUs. The network design includes the use of a [hard swish activation](https://paperswithcode.com/method/hard-swish) and [squeeze-and-excitation](https://paperswithcode.com/method/squeeze-and-excitation-block) modules in the [MBConv blocks](https://paperswithcode.com/method/inverted-residual-block).
The weights from this model were ported from [Tensorflow/Models](https://github.com/tensorflow/models).
A **TResNet** is a variant on a [ResNet](https://paperswithcode.com/method/resnet) that aim to boost accuracy while maintaining GPU training and inference efficiency. They contain several design tricks including a SpaceToDepth stem, [Anti-Alias downsampling](https://paperswithcode.com/method/anti-alias-downsampling), In-Place Activated BatchNorm, Blocks selection and [squeeze-and-excitation layers](https://paperswithcode.com/method/squeeze-and-excitation-block).
The **Vision Transformer** is a model for image classification that employs a Transformer-like architecture over patches of the image. This includes the use of [Multi-Head Attention](https://paperswithcode.com/method/multi-head-attention), [Scaled Dot-Product Attention](https://paperswithcode.com/method/scaled) and other architectural features seen in the [Transformer](https://paperswithcode.com/method/transformer) architecture traditionally used for NLP.
**Wide Residual Networks** are a variant on [ResNets](https://paperswithcode.com/method/resnet) where we decrease depth and increase the width of residual networks. This is achieved through the use of [wide residual blocks](https://paperswithcode.com/method/wide-residual-block).
**Xception** is a convolutional neural network architecture that relies solely on [depthwise separable convolution layers](https://paperswithcode.com/method/depthwise-separable-convolution).
The weights from this model were ported from [Tensorflow/Models](https://github.com/tensorflow/models).
Ross Wightman
4 years ago