Issue with nvdspreprocess Plugin in DeepStream Pipeline for tao_segmentation Example

[john_c]

Please provide complete information as applicable to your setup.

• Hardware Platform (Jetson / GPU) Jetson AGX Orin
• DeepStream Version 7.1
• JetPack Version (valid for Jetson only) 6.2.1
• TensorRT Version 10.3
• NVIDIA GPU Driver Version (valid for GPU only)
• Issue Type( questions, new requirements, bugs) questions

I have some questions about using the nvdspreprocess plugin in a DeepStream pipeline. I’m trying to add nvdspreprocess to the example provided in deepstream_tao_apps/apps/tao_segmentation at master · NVIDIA-AI-IOT/deepstream_tao_apps · GitHub. However, when I set input-tensor-from-meta=1 to enable nvdspreprocess, the pipeline doesn’t work as expected. I haven’t modified libcustom2d_preprocess.so and am using the default version. Below is my configuration file.

config_preprocess.txt (3.0 KB)
pgie_unet_tao_config_yml.txt (2.4 KB)

When I set input-tensor-from-meta=1, I don’t get the mask result, but when I set it to 0, I get the expected result.

[kesong]

Move to DeepStream forum from JPS forum.

[Fiona.Chen]

Seems you are using the YAML configuration deepstream_tao_apps/configs/app/seg_app_unet.yml at master · NVIDIA-AI-IOT/deepstream_tao_apps to run the deepstream_tao_apps/apps/tao_segmentation at master · NVIDIA-AI-IOT/deepstream_tao_apps app.

Since the TAO UNET model size is 320x320, please modify the “width” and “height” in “segvisual” part in deepstream_tao_apps/configs/app/seg_app_unet.yml at master · NVIDIA-AI-IOT/deepstream_tao_apps

segvisual:
  gpu-id: 0
  #The segvisual width and height should be set as the same as the segmentation modle output resolution
  width: 320
  height: 320
  orig_background: 1
  alpha: 0.5