How to get meta when classifier is pgie

[gff1038m]

• Jetson AGX Xavier
• Deepstream 6.0
• JetPack 4.6
• TensorRT 8.0.1
• NVIDIA GPU Driver 32.6.1

Hi, i’m trying to a simple classification pipeline, I need to get labels what my model predicted. In osd_sink_pad_buffer_probe() function i’m trying to get obj_meta_list as written there: Primary classifier class_id=-1 and result_prob=0.58 for all frames and there: Access output of Primary Classifier and get NvDsClassifierMeta. But frame_meta.obj_meta_list return None.

Here is my config:

[property]
process-mode=1 # Process full frames
gpu-id=0
net-scale-factor=1
model-engine-file=classifier.onnx_b1_gpu0_fp32.engine
onnx-file=classifier.onnx
model-color-format=0
network-type=1 # Classifier
network-mode=2 # FP16
#num-detected-classes=2
interval=1
gie-unique-id=1
output-blob-names=predictions/Softmax
classifier-threshold=0.5

And here is code

def osd_sink_pad_buffer_probe(pad,info,u_data):
  gst_buffer = info.get_buffer()
      if not gst_buffer:
          print("Unable to get GstBuffer ")
          return
      batch_meta = pyds.gst_buffer_get_nvds_batch_meta(hash(gst_buffer))
      l_frame = batch_meta.frame_meta_list
      
      while l_frame is not None:
          try:
              frame_meta = pyds.NvDsFrameMeta.cast(l_frame.data)
          except StopIteration:
              break
          
          frame_number = frame_meta.frame_num
          l_obj=frame_meta.obj_meta_list
          print(l_obj) --> It prints None

So how can I get labels that classifier returned?

[yuweiw]

Which model did you use, our demo model or your own model?
You can refer the link below first to get the object:
https://github.com/NVIDIA-AI-IOT/deepstream_python_apps/tree/master/apps/deepstream-test2

[gff1038m]

I use own resnet50

[gff1038m]

Well, I used it as a refer. I did the same, but l_obj is None. Why?

[yuweiw]

It may be your model’s problem. You can verify it by yourself.
1.You can attach your whole code or attach your pipeline graph.
https://forums.developer.nvidia.com/t/deepstream-sdk-faq/80236/10
2.You can modify some code to get the l_obj from our demo code that I attached to see if you can get the value from it.

[gff1038m]

inference_classifier.py (9.8 KB)

Here is the code. I deleted some buisnes logic. It doesn’t tuch any MetaData.

[gff1038m]

Well, I can try your network in my code. It willn’t generate any normal results, but it can create object meta.

[gff1038m]

I used resnet10 with config from deepstream-test3. And l_obj is None

[gff1038m]

I’ve lounched deepstream-test2 and l_obj is not None. But there is a detector. So culd you tell me, is the problem in my pipeline or in way how i’m trying to get l_obj? I can’t understand

[gff1038m]

I’ve modified code and made sgie1 a primary detector (process-mode=1), and I didn’t get any objects. frame_meta.obj_meta_list returned None

[yuweiw]

There is no update from you for a period, assuming this is not an issue anymore.
Hence we are closing this topic. If need further support, please open a new one.
Thanks

You should make sure that your model works well. You can refer the demo below to get the metadata:
https://github.com/NVIDIA-AI-IOT/deepstream_python_apps/blob/master/apps/deepstream-imagedata-multistream/deepstream_imagedata-multistream.py
You can try to use your model instead of our demo model to see if it works well.