vis4d.op.seg.semantic_fpn¶

Semantic FPN Head for segmentation.

Classes

`SemanticFPNHead`([num_classes, in_channels, ...])	SemanticFPNHead used in Panoptic FPN.
`SemanticFPNOut`(outputs)	Output of the SemanticFPN prediction.

class SemanticFPNHead(num_classes=53, in_channels=256, inner_channels=128, start_level=2, end_level=6, dropout_ratio=0.1)[source]¶

SemanticFPNHead used in Panoptic FPN.

Creates an instance of the class.

Parameters:

num_classes (int) – Number of classes. Default: 53.
in_channels (int) – Number of channels in the input feature map.
inner_channels (int) – Number of channels in inner features.
start_level (int) – The start level of the input features used in SemanticFPN.
end_level (int) – The end level of the used features, the end_level-th layer will not be used.
dropout_ratio (float) – The drop ratio of dropout layer. Default: 0.1.

__call__(feats)[source]¶

Type definition for function call.

forward(features)[source]¶

Transforms feature maps and returns segmentation prediction.

Parameters:: features (list[Tensor]) – List of multi-level image features.
Returns:: Segmentation outputs.
Return type:: SemanticFPNOut

init_weights()[source]¶

Initialize weights.

class SemanticFPNOut(outputs: Tensor)[source]¶

Output of the SemanticFPN prediction.

Create new instance of SemanticFPNOut(outputs,)