deeptrees.model

The following flavors of models inputs are possible with DeepTrees: - 4-channel (RGBi). - 5-channel (RGBi+NDVI). See sample config –> https://codebase.helmholtz.cloud/taimur.khan/DeepTrees/-/blob/main/config/train/train_halle.yaml - 7-channel (RGBi+NDVI+GCI+Hue). See sample config –> https://codebase.helmholtz.cloud/taimur.khan/DeepTrees/-/blob/main/config/train/train_halle_7ch.yaml

deeptrees.model.deeptrees_model

deeptrees.model.distance_model

class deeptrees.model.distance_model.DistanceModel(in_channels, architecture='Unet', backbone='resnet18', encoder_weights=None)[source]

Bases: LightningModule

__init__(in_channels, architecture='Unet', backbone='resnet18', encoder_weights=None)[source]

Distance transform model

The model is the second part in the tree crown delineation model.

Parameters:

  • in_channels (int) – Number of input channels

  • architecture (str) – One of ‘Unet, Unet++, Linknet, FPN, PSPNet, PAN, DeepLabV3, DeepLabV3+’

  • backbone (str) – One of the backbones supported by the [pytorch segmentation models package](https://github.com/qubvel/segmentation_models.pytorch)

  • encoder_weights (str | None) – Encoder pretrained weights to load in SMP (e.g. “imagenet”). Use None to skip.

forward(img, mask_and_outline, from_logits=False)[source]

Distance transform forward pass

Parameters:

  • img (torch.Tensor) – Input image

  • mask_and_outline (torch.Tensor) – Tensor containing mask and outlines concatenated in channel dimension, coming from the first sub-network.

  • from_logits (bool) – If set to true, sigmoid activation is applied to the mask_and_outline tensor.

Returns:

Model output of dimension N1HW

deeptrees.model.inference_model

class deeptrees.model.inference_model.InferenceModel(model)[source]

Bases: Module

Just a wrapper to apply the sigmoid activation to mask and outlines during inference..

__init__(model)[source]

Initialize internal Module state, shared by both nn.Module and ScriptModule.

forward(x)[source]

Perform a forward pass through the model. :type x: torch.Tensor :param x: Input tensor to the model. :type x: torch.Tensor

Returns:

If the model output contains two elements, returns a tuple (y, metric) where y is the output tensor with sigmoid activation applied to the first two columns, and metric is the second element of the output. If the model output contains only one element, returns y with sigmoid activation applied to the first two columns.

Return type:

torch.Tensor or tuple

deeptrees.model.segmentation_model

class deeptrees.model.segmentation_model.SegmentationModel(in_channels=4, architecture='Unet', backbone='resnet18', encoder_weights=None)[source]

Bases: LightningModule

__init__(in_channels=4, architecture='Unet', backbone='resnet18', encoder_weights=None)[source]

Segmentation model

A segmentation model which takes an input image and returns a foreground / background mask along with object outlines.

Parameters:

  • in_channels (int) – Number of input channels

  • architecture (str) – One of ‘Unet, Unet++, Linknet, FPN, PSPNet, PAN, DeepLabV3, DeepLabV3+’

  • backbone (str) – One of the backbones supported by the [pytorch segmentation models package](https://github.com/qubvel/segmentation_models.pytorch)

  • encoder_weights (str | None) – Encoder pretrained weights to load in SMP (e.g. “imagenet”). Use None to skip.

  • between. (one means only the mask loss is relevant. Linear in)

forward(x)[source]

Same as torch.nn.Module.forward().

Parameters:

  • *args – Whatever you decide to pass into the forward method.

  • **kwargs – Keyword arguments are also possible.

Returns:

Your model’s output

Contents