Difference between detector_bbox_info and rect_params of object_meta object in pyds

[stct]

• Hardware Platform: Lenovo Legion 7 16IAX7 with a NVIDIA GeForce RTX 3080
• DeepStream Version: 7.07.0.0
• TensorRT Version: 8.6.1.6
• NVIDIA GPU Driver Version (valid for GPU only): 12.2

Description
When extracting bounding box information in a probe using pyds (v. 1.1.11) the object meta object for each detection has the attributes detector_bbox_info.org_bbox_coords and rect_params. From both the width, height, top, right of the bounding box can be extracted. However they seem to differ slightly in value. For example printing the following values for each object meta:

print("\n")
detect_bbox = obj_meta.detector_bbox_info.org_bbox_coords
print(detect_bbox.left, detect_bbox.top, detect_bbox.width, detect_bbox.height)
rect_params = obj_meta.rect_params
print(rect_params.left, rect_params.top, rect_params.width, rect_params.height)

I get the following samle outputs:

1376.25 79.9375 80.75 174.125
1373.23681640625 81.48611450195312 80.74923706054688 173.61602783203125


1497.6875 105.25 73.625 173.0
1493.641357421875 105.43584442138672 74.4700927734375 174.87367248535156


429.0 137.21484375 76.0 171.9140625
431.2253723144531 137.7985076904297 71.65608215332031 170.8712158203125

Why is there a slight difference and which are recommended to use generally?
I also use a tracker element in the Deepstream pipeline. Is this related to this?

[yuweiw]

@stct

I also use a tracker element in the Deepstream pipeline. Is this related to this?

Yes. The more detailed instructions are available in our source code NvDsObjectMeta in the deepstream\sources\includes\nvdsmeta.h. You can refer to that.

[stct]

Thanks for the answer. The answer is now also in the updated documentation: NvDsObjectMeta — Deepstream Deepstream Version: 6.1 documentation