Rotated camera + Counter rotate frames -> No inference on nvinfer

[arivarasan.e]

Please provide complete information as applicable to your setup.

• Hardware Platform (Jetson / GPU) - Jetson Nano
**• DeepStream Version - 6.0 **
• JetPack Version (valid for Jetson only) - 4.6
• TensorRT Version - 8.0
• NVIDIA GPU Driver Version (valid for GPU only)
• Issue Type( questions, new requirements, bugs) - questions

I rotated my camera sensor by 90 degrees to cover the varied heights of persons for my facial recognition project. I, I use flip-method=3 in nvvideoconvert in-order to counter adjust the rotation so that frames reach nvinfer in the expected orientation. nvinfer is a face-detector which is NVIDIA’s FaceDetect model(FaceDetect | NVIDIA NGC). With this rotation + counter rotation, nvinfer couldn’t detect faces.
I’m attaching the code here
inference_module.py (30.7 KB)

Any clue on why nvinfer couldn’t detect after rotation + counter rotation?

[fanzh]

1 which deepstream sample did you refer to?
2 have your code ever detected successfully?
3 can your code detect if no rotation + counter rotation?

[arivarasan.e]

1. It’s based on this open source code - GitHub - riotu-lab/deepstream-facenet: Demo app on using Deepstream 5.0 with Facenet
   2 & 3. Yes, before rotation + counter rotation detector successfully detected faces.

No worries, I have solved the issue.

Root Cause & Fix:
Rotating the frame and resizing it stretches the pixels in a way that it becomes difficult for the detector. I enabled ‘enable-padding’ property in streammux to avoid this. Now the detector can detect even after rotation + counter rotation.

[fanzh]

Glad to know you fixed it, thanks for the update!