Eliminating not complete ojects on border of the frame

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Hi,
I have application that takes images of a area with fast moving objects. Somtimes the objects are not fully in the picture and only part of them are in the image. I want to use “roi-top-offset” and “roi-bottom-offset” config in the detector config file (YOLO) in deepstream to be able to eliminate considering the partial objects. The partial objects all have center of the BB close to bottom or top. There are some full objects that have BB close to the edge, howevve the center of BB is not close to the edge. That is the difference. If I use roi-top-offset will that also eliminate the detected objects with BB that have its center away from the offset region, but a portion of BB is in the offset region?
Thanks!

2

Moving to DeepStream forum.

4

NO. Instead of eliminating the object, it will correct the coordinate of the bbox.
You can refer to our source code in sources\gst-plugins\gst-nvinfer\gstnvinfer_meta_utils.cpp.

void
attach_metadata_detector (GstNvInfer * nvinfer, GstMiniObject * tensor_out_object,
    GstNvInferFrame & frame, NvDsInferDetectionOutput & detection_output, float segmentationThreshold)
5

How I can perform the task of elimination of partial objects from the detection outputs? I want the objects that are only partial and center of the BB is near the edge of the image, be eliminated, and not be detected and outputed as an object.

6

You can modify the source code I attached to meet your needs. This is where you decide which objects you want to eliminate.

7

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

8

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.