Custom Semantic Segmentation Parameter in nvinfer Configuration File

• Hardware Platform (Jetson / GPU) : Jetson Orin Nano
• DeepStream Version : 7.1
• JetPack Version (valid for Jetson only) : 6.2
• TensorRT Version: **10.3.0.30
• NVIDIA GPU Driver Version (valid for GPU only) : 540.4.0

As per the nvinfer Documentation:

  • parse-classifier-func-name: is for classification
  • parse-bbox-instance-mask-func-name: is for instance segmentation
  • parse-bbox-func-name: is for object detection

However, I cannot find any parameter specifically for semantic segmentation. Despite this, I see files for creating custom parser libraries for various tasks, including semantic segmentation, in the following directory:

/opt/nvidia/deepstream/deepstream-7.1/sources/libs/

Files:

  • Makefile
  • nvdsinfer_custombboxparser.cpp
  • nvdsinfer_customclassifierparser.cpp
  • nvdsinfer_customsegmentationparser.cpp
  • README

If this is not yet implemented, could you please let me know if it will be included in the next release?

We currently do not support customizing this parsing. But you can refer to our soure code to learn how to attach the segmentation data to the metadata.

deepstream\sources\gst-plugins\gst-nvinfer\gstnvinfer_meta_utils.cpp
void
attach_metadata_segmentation (GstNvInfer * nvinfer, GstMiniObject * tensor_out_object,
    GstNvInferFrame & frame, NvDsInferSegmentationOutput & segmentation_output)
{