Please provide complete information as applicable to your setup.
Hardware Platform:- dGPU
DeepStream Version :- 6.2
I have this pipeline
elements_before_streammux → streammux → pgie (face_detector) → tracker → sgie (face recognizer) → sink
I want to transform it to
elements_before_streammux → streammux → pgie (face_detector) → tracker → sgie1 (person_detector) → sgie2 (face recognizer) → sink
(this is exactly what I want for my use case)
I want to make the sgie2 (face recognizer) work on pgie(face detector) and sgie1 on full frame. If I change the process-mode of sgie1 to work on full frame, will I still be able to get the obj_meta->parent pointer (object_meta of sgie1 → parent (pgie) ) ?? I want to be able to use the obj_meta->parent pointer.
Is that possible? If yes, How to do that ?
Your pipeline looks strange. Do you mean you want the person_detector’s result to be connected to the face_detector and face_recognizer 's results?
Why don’t you use the pipeline of “person_detector(PGIE)”->“face_detector(SGIE1)”->“face_recognizer(SGIE2)”?
Yes. I want to connect person detector after face detector. I tried the other way around, but face detection is primary task and If I connect person_detector->face_detector, I can loose faces because person_detector is not 100% accurate. I’m doing that because, I dont want to loose face detection.
I want to detect a person only if a face is detected.
Will your face_detector output the bbox for face or the bbox for the whole person?
face detector will output the bbox of face of a person
The face bbox is small, it will not be big enough for oerson_detector to find a whole person. So there is no connection between the person_detector and face_detector if you use the pipeline as face_detector->person_detector. The reasonable way is to improve the accuracy of the person_detector and use the correct pipeline.
okay, then we will go with this correct approach only. Thanks.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.