How to make a parses function for my regression model

[ayanasser]

@bcao

, but seems the model’s output is not attached to gst_buffer or the model’s output is empty

Is this because there’s no parsing function could parse this output?
or NvDSTensorMeta does not need parse function ?

[ayanasser]

@bcao

But you need to make sure the input stream inside DS package can be used for your model

and could you please explain more what is this point mean?
The model takes 1 frame as input with shape (3x720x1280), Is this input could not work with the deepstream 5 ?

[bcao]

There are some issue in nvinfer, just update a rough WAR for you, you can install the probe on src pad of nvinfer plugin instead of sink pad of osd as following:

osdsinkpad = nvosd.get_static_pad("sink")
chage to 
osdsinkpad = pgie.get_static_pad("src")

Also for how to parse NvDsInferTensorMeta, you need to refer deepstream_python_apps/ssd_parser.py at 5cb4cb8be92e079acd07d911d265946580ea81cd · NVIDIA-AI-IOT/deepstream_python_apps · GitHub for how to access the out_buf_ptrs_host

I would suggest to create a new topic for the NvDsInferTensorMeta missed issue if you would like to install the probe on sink pad of osd plugin.