Help required for visual changenet segmentation model

Please provide the following information when requesting support.

• Hardware (T4/V100/Xavier/Nano/etc) : A6000
• Network Type (Detectnet_v2/Faster_rcnn/Yolo_v4/LPRnet/Mask_rcnn/Classification/etc) : Changenet
• TLT Version (Please run “tlt info --verbose” and share “docker_tag” here) : 5.3.0
• Training spec file(If have, please share here)
• How to reproduce the issue ? (This is for errors. Please share the command line and the detailed log here.)

I have a trained a visual changenet segmentation model using custom dataset & deployed it on triton. I am unable to do postprocessing of the results like how to extract conf score , class , segmentation mask details … is there any sample code available for me to take a look.
output [
{
name: “output_final”
data_type: TYPE_FP32
dims: [ 2, 256, 256 ]
},
{
name: “output0”
data_type: TYPE_FP32
dims: [ 2, 16, 16 ]
},
{
name: “output1”
data_type: TYPE_FP32
dims: [ 2, 16, 16 ]
},
{
name: “output2”
data_type: TYPE_FP32
dims: [ 2, 32, 32 ]
},
{
name: “output3”
data_type: TYPE_FP32
dims: [ 2, 64, 64 ]
}
]

You can refer to

  1. triton-apps: tao-toolkit-triton-apps/tao_triton/python/postprocessing/visual_changenet_postprocessor.py at main · NVIDIA-AI-IOT/tao-toolkit-triton-apps · GitHub

or

  1. tao-deploy:
    tao_deploy/nvidia_tao_deploy/cv/visual_changenet/segmentation/utils.py at 31c7e0ed3fe48942c254b3b85517e7418eea17b3 · NVIDIA/tao_deploy · GitHub,
    tao_deploy/nvidia_tao_deploy/cv/visual_changenet/scripts/inference.py at 31c7e0ed3fe48942c254b3b85517e7418eea17b3 · NVIDIA/tao_deploy · GitHub.
1 Like

@Morganh : Thanks for reference. I have couple of question as am little new to this and trying to understand.

  1. Let’s say i have passed a golden image and a image , i get a results the in output_final with shape (2,256,256) where first channel consist of pixel wise probabilities of change and similarly second channel consist of pixel wise probabilities of no change. & used argmax to calculate the max value across the pixels , This gives me a binary mask. is my understanding correct. ? and also how do i calculate the confidence score ?
  2. Is there any useful information which i can get it through other outputs like output0,output1,…

Yes,

There is not confidence score. But you can get the array from tao-toolkit-triton-apps/tao_triton/python/postprocessing/visual_changenet_postprocessor.py at main · NVIDIA-AI-IOT/tao-toolkit-triton-apps · GitHub and set a threshold for it.

You can draw their output. Similar to tao-toolkit-triton-apps/tao_triton/python/postprocessing/visual_changenet_postprocessor.py at main · NVIDIA-AI-IOT/tao-toolkit-triton-apps · GitHub. More info about the outputs can be found in tao_pytorch_backend/nvidia_tao_pytorch/cv/visual_changenet/segmentation/models/changenet.py at dc07b02eb78c2eb868315107892b466496e55a0f · NVIDIA/tao_pytorch_backend · GitHub.

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.