How to fetch the frame meta data in custom bbox-parsing (parse-bbox-func-name)?

290844930 · January 4, 2021, 7:57am

Please provide complete information as applicable to your setup.

• Hardware Platform (Jetson / GPU)
GPU
• DeepStream Version
.5.0
• JetPack Version (valid for Jetson only)
• TensorRT Version
7.1
• NVIDIA GPU Driver Version (valid for GPU only)
440
• Issue Type( questions, new requirements, bugs)
question

For Yolov5, there is a custom bbox-parsing configured in config file. I hope the info from frame meta data could be accessed in some way. Any suggestion for that? for example, the scale ratio or dimension of frame input?

bcao · January 4, 2021, 9:59am

Yeah, we can access the model output layer info and network info, please check nvdsinfer_custombboxparser.cpp, there are many customized post process parser for your reference.

290844930 · January 4, 2021, 10:05am

yes, I have viewed these source codes, but didnt find some meta data like width/height info of the original input frame, only the network width/height. could you help me further?

bcao · January 4, 2021, 10:58am

Right, we don’t provide width/height info of the original input frame in customized post process parser function. But you can access the info in gstnvinfer_meta_utils.cpp -> attach_metadata_detector.