How to add numpy images in nvdsBufSurface deepstream

[phammanhtuongtfos]

• Hardware Platform: Tesla T4
• DeepStream Version : 5.1
• TensorRT Version : 7.2.2.3
• NVIDIA GPU Driver Version: 460.32.03
• Issue Type: Questions

After my detection model is executing, I’m modifying dsexample plugin for resize and rotate objects detect and then i using classification model. Now, I’m overwrite output of my dsexample plugin on source frame in nvdsBufSurface and update coordinate in NvOSD_RectParams to use for classification models, but that would make a bad videos output. How i can create new NvDsBatchMeta with objects detect are processing in dsexample as image input.
Does anyone have a better solution?
Thanks!

[bcao]

Hey, What’s your pipeline?

[phammanhtuongtfos]

My pipeline:
source ( videos) → h264parser → decoder → streammux → pgie → tracker → nvvidconv → dsexample → sgie → nvosd → nvvidconv2 → videoconverter → x246encode → qtmux → sink

[bcao]

OK, seems you are doing some cumtomized preprocess here, also is it possible to do that in the sgie?

[phammanhtuongtfos]

that sound good, do you have some example code for customized in the sgie? can you share it for me, pleasse!

[bcao]

You need to do cumtomized preprocess in gstnvinfer.cpp → gst_nvinfer_process_objects
I think there are some similar topics in the forum