Secondary detector input size

When we pass primary detector detected object to secondary object classification then which size is considered to crop image and pass to second detector.
(crop primary detected objects for secondary inferencing)

Size at Muxer -> primary object inferencing in scale-> secondary inference on detected object.

The crop size is the coordinate of the BBOX detected by primary GIE. The cropped image will be resized in the secondary GIE to the input size of the model.

1 Like

Hi Thanks for the reply. But its not clear to me.

I give you one example so that I can get more details that I want.


This data pass to PGIE
Let say PGIE do the inference on network size
width = 416
. So my question is data would be crop from streammux size then resize to SGIE input size of data would be crop in network size i.e. 416X416 then resize to SGIE input size.

Following your example, suppose PGIE detected an object with a BBOX coordinate, then SGIE will use the BBOX coordinate to crop the image from the 1920x1080 image buffer in streammux, then resize these cropped images to network size of the SGIE and do the SGIE inference

understood. Thanks for reply.