Image Classification (TF2) Deesptream/Triton Server Config File

I was be able to train, export , create engine trt file on Triton-Server
But I’m missing some configuration on deepstream preprocessing.

All tests are successful including convert engine and visualizing TensorRT inferences.
But It’s not working (0 results after inference) when use on Deepstream/Triton-Server.
I think is missing some configuration on deesptream preprocessing.

During Export of Classification TF1 tao generated triton-sever and deepstream-config, but Classification TF2 doesnt generate it.

SGIE Deepstream Triton Config

infer_config {
  unique_id: 6
  max_batch_size: 50
  
  backend {
    triton {
      model_name: "vehicletypenet"
      version: -1
      grpc {
        url: "0.0.0.0:8001"
        enable_cuda_buffer_sharing: true
      }
    }
  }

  preprocess {
    network_format: IMAGE_FORMAT_BGR
    tensor_order: TENSOR_ORDER_LINEAR
    maintain_aspect_ratio: 0
    frame_scaling_hw: FRAME_SCALING_HW_DEFAULT
    frame_scaling_filter: 1
    normalize {
      scale_factor: 0.017507
      channel_offsets: [123.675,116.280,103.53]
      #mean_file: "mean_vehiclemake.ppm"
    }
  }


  postprocess {
    labelfile_path: "label_vehicletype.txt"
    classification {
      threshold: 0.30
    }
  }
}

input_control {
  process_mode: PROCESS_MODE_CLIP_OBJECTS
  operate_on_gie_id: 1
  operate_on_class_ids: [0,3]
  interval: 0
  async_mode: true
  object_control {
    bbox_filter {
      min_width: 128
      min_height: 128
    }
  }
}

Triton-Config


name: "vehicletypenet"
platform: "tensorrt_plan"
max_batch_size: 128 
default_model_filename: "efficientnet-b0.fp16x128_new.engine"
input [
  {
    name: "input:0"
    data_type: TYPE_FP32
    format: FORMAT_NCHW
    dims: [3, 256, 256]
  }
]
output [
  {
    name: "Identity:0"
    data_type: TYPE_FP32
    dims: [12]
  }
]
instance_group [
  {
    kind: KIND_GPU
    count: 1
    gpus: 0
  }
]

Found the issue.

Preprocessing:
Changed

from :

preprocessing_mode: “caffe”
net-scale-factor=1.0
offsets=B;G;R
model-color-format=1

To:

preprocess_mode: ‘torch’
scale_factor: 0.017507
channel_offsets: [123.675,116.280,103.53]
model-color-format=1

https://docs.nvidia.com/tao/tao-toolkit/text/ds_tao/classification_ds.html#deploying-to-deepstream-classification

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