thelper.nn.segmentation package

Neural network and model package for segemnation.

This package contains classes that define blocks and modules used in various neural network for segmentation

Submodules

thelper.nn.segmentation.base module

class thelper.nn.segmentation.base.SegmModelBase(task, pretrained=False)

Bases: thelper.nn.utils.Module

Base wrapper class for specialized segmentation models.

__init__(task, pretrained=False)

Note

forward(x)

Transforms an input tensor in order to generate a prediction.

in_channels = None

model_cls = None

set_task(task)

Adapts the model to support a new task, replacing layers if needed.

thelper.nn.segmentation.deeplabv3 module

class thelper.nn.segmentation.deeplabv3.DeepLabV3ResNet101(task, pretrained=False)

Bases: thelper.nn.segmentation.base.SegmModelBase

This class is a thin wrapper for torchvision.models.segmentation.deeplabv3_resnet101() (torchvision > 0.6).

Note

Contributed by Mario Beaulieu <mario.beaulieu@crim.ca>.

See also

Liang-Chieh et al., Rethinking Atrous Convolution for Semantic Image Segmentation [arXiv], 2017.

__init__(task, pretrained=False)

class thelper.nn.segmentation.deeplabv3.DeepLabV3ResNet50(task, pretrained=False)

Bases: thelper.nn.segmentation.base.SegmModelBase

This class is a thin wrapper for torchvision.models.segmentation.deeplabv3_resnet101() (torchvision > 0.6).

Note

Contributed by Mario Beaulieu <mario.beaulieu@crim.ca>.

See also

Liang-Chieh et al., Rethinking Atrous Convolution for Semantic Image Segmentation [arXiv], 2017.

__init__(task, pretrained=False)

thelper.nn.segmentation.fcn module

class thelper.nn.segmentation.fcn.FCNResNet101(task, pretrained=False)

Bases: thelper.nn.segmentation.base.SegmModelBase

This class is a thin wrapper for torchvision.models.segmentation.fcn_resnet50() (torchvision > 0.6).

Note

Contributed by Mario Beaulieu <mario.beaulieu@crim.ca>.

See also

Liang-Chieh et al., Rethinking Atrous Convolution for Semantic Image Segmentation [arXiv], 2017.

__init__(task, pretrained=False)

class thelper.nn.segmentation.fcn.FCNResNet50(task, pretrained=False)

Bases: thelper.nn.segmentation.base.SegmModelBase

This class is a thin wrapper for torchvision.models.segmentation.fcn_resnet50() (torchvision > 0.6).

Note

Contributed by Mario Beaulieu <mario.beaulieu@crim.ca>.

See also

Liang-Chieh et al., Rethinking Atrous Convolution for Semantic Image Segmentation [arXiv], 2017.

__init__(task, pretrained=False)