# SegmentationParser SegmentationParser node takes the raw output of a segmentation NN and applies per-pixel argmax to convert the raw output tensor into a single-channel segmentation mask, which is then output as a [SegmentationMask](https://docs.luxonis.com/software-v3/depthai/depthai-components/messages/segmentation_mask.md) message. Some segmentation NNs output the segmentation mask directly as a single-channel output, in which case the SegmentationParser node can be used to simply convert the raw output tensor into a SegmentationMask message without applying argmax. ## How to place it #### Python ```python pipeline = dai.Pipeline() segmentation = pipeline.create(dai.node.SegmentationParser) ``` #### C++ ```cpp dai::Pipeline pipeline; auto segmentation = pipeline.create(); ``` ## Inputs and Outputs ## Usage #### Python ```python model = dai.NNModelDescription("luxonis/deeplab-v3-plus:512x512") pipeline = dai.Pipeline() camera = pipeline.create(dai.node.Camera).build() nn = pipeline.create(dai.node.NeuralNetwork).build(camera, model) parser = pipeline.create(dai.node.SegmentationParser).build(nn.out, model ) ``` #### C++ ```cpp dai::NNModelDescription model("luxonis/deeplab-v3-plus:512x512"); dai::Pipeline pipeline; auto camera = pipeline.create()->build(); auto nn = pipeline.create()->build(camera, model); auto parser = pipeline.create()->build(nn->out, model); ``` ## Examples of functionality * [Semantic segmentation (Python)](https://github.com/luxonis/depthai-core/blob/main/examples/python/Segmentation/semantic_segmentation.py) - Run a DeepLabv3+ segmentation model and display the segmentation masks on a video stream. * [Segmentation examples (C++)](https://github.com/luxonis/depthai-core/tree/main/examples/cpp/Segmentation) - Browse the C++ segmentation examples in `depthai-core`. ## Reference ### dai::node::SegmentationParser Kind: class SegmentationParser node. Parses raw segmentation output from segmentation neural networks into a dai::SegmentationMask datatype. The parser supports two output model types: Single-channel output where the model argmaxes the class probabilities internally and outputs a single channel mask with class indices.; Multi-channel output where each channel corresponds to the probability map for a specific class. The parser will perform argmax across channels to generate the final mask. The parser can be configured to treat the first class (index 0) as the background class, which will be ignored in the final segmentation mask. warning: Only OAK4 supports running SegmentationParser on device. On other platforms, the node will automatically switch to host execution. #### std::variant< NNModelDescription , NNArchive , std::string > Model Kind: typedef #### std::shared_ptr< SegmentationParserConfig > initialConfig Kind: variable Initial config to use when parsing segmentation masks. #### Input input Kind: variable Input NN results with segmentation data to parser #### Input inputConfig Kind: variable Input SegmentationParserConfig message with ability to modify parameters in runtime. #### Output out Kind: variable Outputs segmentation mask #### ~SegmentationParser() Kind: function #### SegmentationParser() Kind: function #### SegmentationParser(std::unique_ptr< Properties > props) Kind: function #### std::shared_ptr< SegmentationParser > build(Node::Output & nnInput, const Model & model) Kind: function Build SegmentationParser node. parameters: nnInput: Output to link; model: Neural network model #### std::shared_ptr< SegmentationParser > build(Node::Output & nnInput, const dai::nn_archive::v1::Head & head) Kind: function Build SegmentationParser node with the specific head from NNArchive . Useful when model outputs multiple segmentation heads. parameters: nnInput: Output to link; head: Specific head from NNArchive to use for this parser #### void setNNArchive(const NNArchive & nnArchive) Kind: function Set NNArchive for this Node . parameters: nnArchive: NNArchive to set #### void setNNArchiveHead(const dai::nn_archive::v1::Head & head) Kind: function Set NNArchive head for this Node . parameters: head: NNArchive head to set #### void setLabels(const std::vector< std::string > & labels) Kind: function Sets the class labels associated with the segmentation mask. The label at index $i$ in the labels vector corresponds to the value $i$ in the segmentation mask data array. parameters: labels: Vector of class labels #### std::vector< std::string > getLabels() Kind: function Returns the class labels associated with the segmentation mask. #### void setBackgroundClass(bool backgroundClass) Kind: function Sets whether the first class (index 0) is considered the background class. If true, the pixels classified as index 0 will be treated as background. parameters: backgroundClass: Boolean indicating if the first class is the background class note: Only applicable if the number of classes is greater than 1 and the output classes are not in a single layer (eg. classesInOneLayer = false). #### bool getBackgroundClass() Kind: function Gets whether the first class (index 0) is considered the background class. #### void setRunOnHost(bool runOnHost) Kind: function Specify whether to run on host or device By default, the node will run on device. #### bool runOnHost() Kind: function Check if the node is set to run on host #### void run() Kind: function #### void buildInternal() Kind: function Function called from within the create function to build the node. This function is useful for initialization, setting up inputs and outputs = stuff that cannot be perform in the constuctor. #### DeviceNodeCRTP() Kind: function #### DeviceNodeCRTP(const std::shared_ptr< Device > & device) Kind: function #### DeviceNodeCRTP(std::unique_ptr< Properties > props) Kind: function #### DeviceNodeCRTP(std::unique_ptr< Properties > props, bool confMode) Kind: function #### DeviceNodeCRTP(const std::shared_ptr< Device > & device, std::unique_ptr< Properties > props, bool confMode) Kind: function ### Need assistance? Head over to [Discussion Forum](https://discuss.luxonis.com/) for technical support or any other questions you might have.