Composition of NvDsInferLayerInfo list

[sujin.kim]

Please provide complete information as applicable to your setup.

• Hardware Platform (Jetson / GPU) GPU
• DeepStream Version 6.2
• JetPack Version (valid for Jetson only)
• TensorRT Version 8.5.2-1+cuda11.8
• NVIDIA GPU Driver Version (valid for GPU only) 525.85.12
• Issue Type( questions, new requirements, bugs) questions

For YOLO-based object detection,
I set ‘parse-bbox-func-names’ using 3rd party source code.
And I could find some ambiguous things from external code, regarding output parsing-function.
One of inputs for parsing-function is std::vector,
and some predefined indexing is used in parsing-function.
for example, index ‘0’, for list of bbox’s position and size,
index ‘1’, for list of scores,
index ‘2’, for list of classes.
who defined vector-form of NvDsInferLayerInfo?
This is my first question.
AS I know, user should set ‘network-type’ by predefined integer value.
So when ‘network-type’ is 0, then will nvinfer (in deepstream) automatically assign
values for vector of NvDsInferLayerInfo, like above?
If it’s correct, When user executes detection model that isn’t YOLO,
similar indexing from std::vector argument will work well?
This is my 2nd question.

[junshengy]

@sujin.kim

AS I know, user should set ‘network-type’ by predefined integer value.

In fact, these index depend on the model.
Below is the output of yolov8n. As you described above, index ‘0’, for list of bbox’s position and size.

Alternatively, you can pass layerName. Here is an example using layerName:

NVIDIA-AI-IOT/deepstream_tao_apps/blob/master/post_processor/nvdsinfer_custombboxparser_tao.cpp#L352

@sujin.kim

If it’s correct, When user executes detection model that isn’t YOLO,
similar indexing from std::vector argument will work well?

As mentioned above, using layername is a better choice, index is not a good choice.
Unless you know the layer corresponding to the index.

In general, std::vector<NvDsInferLayerInfo> const& outputLayersInfo represents the output layer of the model