Segmentation¶
Below you will find all the latest image segmentation models. To get a list of specific model names that are available programmatically, call the pywick.models.model_utils.get_supported_models(...)
method.
To load one of these models with your own number of classes you have two options:
1. You can always load the model directly from the API. Most models allow you to customize number of classes as well as pretrained options.
2. You can use the pywick.models.model_utils.get_model(...)
method and pass the name of the model
that you want as a string. Note: Some models allow you to customize additional parameters. You can take a look at the pywick.models.model_utils.get_model(...)
method
or at the definition of the model to see what’s possible. pywick.models.model_utils.get_model(...)
takes in a **kwargs
argument that you can populate with whatever parameters you’d like
to pass to the model you are creating.
BiSeNet¶
Implementation of BiSeNet: Bilateral Segmentation Network for Real-time Semantic Segmentation
DANet¶
Implementation of Dual Attention Network for Scene Segmentation
-
class
pywick.models.segmentation.danet.
DANet
(num_classes, pretrained=True, backbone='resnet101', aux=False, **kwargs)[source]¶ Pyramid Scene Parsing Network
- nclass : int
- Number of categories for the training dataset.
- backbone : string
- Pre-trained dilated backbone network type (default:’resnet50’; ‘resnet50’, ‘resnet101’ or ‘resnet152’).
- norm_layer : object
- Normalization layer used in backbone network (default:
mxnet.gluon.nn.BatchNorm
; for Synchronized Cross-GPU BachNormalization). - aux : bool
- Auxiliary loss.
- Reference:
- Jun Fu, Jing Liu, Haijie Tian, Yong Li, Yongjun Bao, Zhiwei Fang,and Hanqing Lu. “Dual Attention Network for Scene Segmentation.” CVPR, 2019
EmaNet¶
DenseASPP¶
Implementation of DenseASPP for Semantic Segmentation in Street Scenes
Deeplab V2 Resnet¶
DeepLab v2 - DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs <https://arxiv.org/abs/1606.00915>`_
Deeplab V3¶
DeepLab v3 - Rethinking Atrous Convolution for Semantic Image Segmentation
Deeplab V3+¶
DeepLab v3+ Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation
DRNNet¶
Implementation of Dilated Residual Networks
DUC, HDC¶
Implementation of: Understanding Convolution for Semantic Segmentation
DUNet¶
Implementation of Decoders Matter for Semantic Segmentation: Data-Dependent Decoding Enables Flexible Feature Aggregation
-
class
pywick.models.segmentation.dunet.
DUNet
(num_classes, pretrained=True, backbone='resnet101', aux=False, **kwargs)[source]¶ Decoders Matter for Semantic Segmentation
- Reference:
- Zhi Tian, Tong He, Chunhua Shen, and Youliang Yan. “Decoders Matter for Semantic Segmentation: Data-Dependent Decoding Enables Flexible Feature Aggregation.” CVPR, 2019
ENet¶
Implementation of ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation
-
class
pywick.models.segmentation.enet.
ENet
(num_classes, encoder_relu=False, decoder_relu=True, **kwargs)[source]¶ Generate the ENet model.
Parameters: - num_classes – (int): the number of classes to segment.
- encoder_relu – (bool, optional): When
True
ReLU is used as the activation function in the encoder blocks/layers; otherwise, PReLU is used. Default: False. - decoder_relu – (bool, optional): When
True
ReLU is used as the activation function in the decoder blocks/layers; otherwise, PReLU is used. Default: True.
FCN8 VGG¶
Implementation of Fully Convolutional Networks for Semantic Segmentation
FCN16 VGG¶
Implementation of Fully Convolutional Networks for Semantic Segmentation
FCN32 VGG¶
Implementation of Fully Convolutional Networks for Semantic Segmentation
FRRN¶
Implementation of Full Resolution Residual Networks for Semantic Segmentation in Street Scenes
-
class
pywick.models.segmentation.frrn1.
frrn
(num_classes=21, model_type=None, **kwargs)[source]¶ Full Resolution Residual Networks for Semantic Segmentation URL: https://arxiv.org/abs/1611.08323
References: 1) Original Author’s code: https://github.com/TobyPDE/FRRN 2) TF implementation by @kiwonjoon: https://github.com/hiwonjoon/tf-frrn
FusionNet¶
Implementation of FusionNet: A deep fully residual convolutional neural network for image segmentation in connectomics
GALDNet¶
GCN Densenet¶
Implementation of Large Kernel Matters with Densenet backend
GCN NASNet¶
Implementation of Large Kernel Matters with NASNet backend
GCN PSP¶
Implementation of Large Kernel Matters with PSP backend
GCN Resnet¶
Implementation of Large Kernel Matters with Resnet backend.
GCN ResNext¶
Implementation of Large Kernel Matters with Resnext backend
Linknet¶
Implementation of LinkNet: Exploiting Encoder Representations for Efficient Semantic Segmentation
-
class
pywick.models.segmentation.mnas_linknets.linknet.
LinkCeption
(num_classes, pretrained=True, num_channels=3, is_deconv=False, decoder_kernel_size=4, **_)[source]¶
-
class
pywick.models.segmentation.mnas_linknets.linknet.
LinkDenseNet121
(num_classes, pretrained=True, num_channels=3, is_deconv=False, decoder_kernel_size=4, **kwargs)[source]¶
-
class
pywick.models.segmentation.mnas_linknets.linknet.
LinkDenseNet161
(num_classes, pretrained=True, num_channels=3, is_deconv=False, decoder_kernel_size=4, **kwargs)[source]¶
-
class
pywick.models.segmentation.mnas_linknets.linknet.
LinkInceptionResNet
(num_classes, pretrained=True, num_channels=3, is_deconv=False, decoder_kernel_size=3, **kwargs)[source]¶
-
class
pywick.models.segmentation.mnas_linknets.linknet.
LinkNet18
(num_classes, pretrained=True, num_channels=3, is_deconv=False, decoder_kernel_size=4, **kwargs)[source]¶
-
class
pywick.models.segmentation.mnas_linknets.linknet.
LinkNet34
(num_classes, pretrained=True, num_channels=3, is_deconv=False, decoder_kernel_size=4, **kwargs)[source]¶
-
class
pywick.models.segmentation.mnas_linknets.linknet.
LinkNet50
(num_classes, pretrained=True, num_channels=3, is_deconv=False, decoder_kernel_size=4, **kwargs)[source]¶
-
class
pywick.models.segmentation.mnas_linknets.linknet.
LinkNet101
(num_classes, pretrained=True, num_channels=3, is_deconv=False, decoder_kernel_size=4, **kwargs)[source]¶
-
class
pywick.models.segmentation.mnas_linknets.linknet.
LinkNet152
(num_classes, pretrained=True, num_channels=3, is_deconv=False, decoder_kernel_size=3, **kwargs)[source]¶
OCNet¶
Implementation of OCNet: Object Context Network for Scene Parsing
-
class
pywick.models.segmentation.ocnet.
OCNet
(num_classes, pretrained=True, backbone='resnet101', oc_arch='base', aux=False, **kwargs)[source]¶ - nclass : int
- Number of categories for the training dataset.
- backbone : string
- Pre-trained dilated backbone network type (default:’resnet50’; ‘resnet50’, ‘resnet101’ or ‘resnet152’).
- norm_layer : object
- Normalization layer used in backbone network (default:
nn.BatchNorm
; for Synchronized Cross-GPU BachNormalization). - aux : bool
- Auxiliary loss.
- Reference:
- Yuhui Yuan, Jingdong Wang. “OCNet: Object Context Network for Scene Parsing.” arXiv preprint arXiv:1809.00916 (2018).
RefineNet¶
Implementation of RefineNet: Multi-Path Refinement Networks for High-Resolution Semantic Segmentation.
PSP¶
Implementation of Pyramid Scene Parsing Network
SegNet¶
Implementation of Segnet: A deep convolutional encoder-decoder architecture for image segmentation
Tiramisu¶
Implementation of The One Hundred Layers Tiramisu: Fully Convolutional DenseNets for Semantic Segmentation
Unet¶
Implementation of U-net: Convolutional networks for biomedical image segmentation
Unet Sized¶
Implementation of U-net: Convolutional networks for biomedical image segmentation
Unet Dilated¶
Implementation of U-net: Convolutional networks for biomedical image segmentation with dilation convolution operation
Unet Residual¶
Implementation of U-net: Convolutional networks for biomedical image segmentation
Unet_stack¶
Implementation of stacked U-net: Convolutional networks for biomedical image segmentation