Drawing Bounding Boxes



Bounding box is by default shown in the deepstream SDK.
For example:

deepstream-app -c samples/configs/deepstream-app/source30_1080p_dec_infer-resnet_tiled_display_int8.txt


I’m using your sample models ‘resnet10.caffemodel’ I wanted to know where do you actually draw the bounding boxes?

nvdsosd plugin draws the bounding boxes

Hi the reason why I’m asking this is that I’m running a face detection model on it and wanted to apply Gaussian blur on the respective frame. If somehow I could get access to the image frame I could pixelate the face detected ROI.

Coming back to your answer there is this osd header file ‘deepstream_osd.h’ that contains some structs ‘NvDsOSDConfig’ and ‘NvDsOSDBin’ but I don’t see where the actual frame is so I can apply the obfuscation on it.

Is it at Gstreamer level. How can I get access to the respective image frames?

Yes there is this app ‘https://github.com/NVIDIA-AI-IOT/redaction_with_deepstream/’ but it only makes a patch on the frame through a plugin. I want to apply Gaussian kernel on the image for that I need to get the hold of the image.

What is the location of that plugin ? I can’t see that file in my deepstream directory

You can refer to dsexample sample to access image frame and apply Gaussian blur
Another sample: https://github.com/NVIDIA-AI-IOT/Redaction_with_DeepStream
Redaction app will apply a solid mask for face (not exactly Gaussian blur)