Hello, I suppose that I have a pose estimation model. My pipeline is that after detected pose with pose estimation model, I use the additional human classifier to check the low confidence score threshold wherether the detected object is person or not. I use deepstream to accelerate my pipeline. So how can I use nvdsinfer plugin to construct my pipeline. Actually The pipeline without Human classifier works fine for me. But I dont know how to put it to my pipeline. Thanks alot.
How to Run a classifier with only low confidence detection from nvdsinfer
Can you tell more details about this human classifier model? What are the input layers and output layers? For example, for yolo models, we know it has only one input layer which is a RGB image.
Actually The input of the Human classifier model is the bounding box cropped by using pose estimated from Human pose estimation model. My pipeline is as follow:
so suppose that our human pose estimation outputs 5 people have the confidence score: [0.9, 0.8, 0.7, 0.6, 0.1]. I worry that the person which has score of 0.1 is the false positive, So I crop the bounding box of this person using detected poses then pass it though the image classifier to check wherether it is false positive or not.
So you can use Human pose estimation model as PGIE and use the Human classifier model as SGIE. There are samples for PGIE+SGIE : /opt/nvidia/deepstream/deepstream/sources/apps/sample_apps/deepstream-test2
Thank for you quick reply. But the thing is that I want to keep all the person poses with high confidence score. So how can I do that? I mean the human classifier acts as the validate module only.
What do you mean by “keep all the person poses with high confidence score”? Who will remove the poses with low scores? I don’t think gst-nvinfer does this.
after the pose estimation model, I will attach the detected poses to the object meta data for the downstream plugin. Then I use the Human classifier as a filter, but that filter will run only on the low confidence score poses.
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.
As you said the filter runs only on the low confidence score poses, will it remove or delete the object meta data with the detected poses you attached? If not, the metadata is always there, the thing you attached is always there. Everything is under your control. DeepStream is only a SDK.
thank you for your support.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.