Savant/samples/multiple_rtsp/demo.yml at develop · Bipolar-Factory/Savant · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
# module name, required
name: ${oc.env:MODULE_NAME, 'demo'}

# base module parameters
parameters:
  # pipeline processing frame parameters
  frame:
    width: 1280
    height: 720
  output_frame:
    codec: ${oc.env:CODEC, 'raw-rgba'}
  # PyFunc for drawing on frames (default implementation)
  draw_func: {}

# pipeline definition
pipeline:
  # source definition is skipped, zeromq source is used by default to connect with source adapters

  # define pipeline's main elements
  elements:
    # primary detector element, inference is provided by the nvinfer Deepstream element
    # model type is detector (other available types are: classifier, custom)
    - element: nvinfer@detector
      # Model's name in the pipeline, mandatory
      name: peoplenet
      # model definition
      model:
        # format of the provided model file
        format: etlt
        # remote storage where the model files can be found
        # skip if providing model files locally
        remote:
          url: s3://savant-data/models/peoplenet/peoplenet_pruned_v2.0.zip
          checksum_url: s3://savant-data/models/peoplenet/peoplenet_pruned_v2.0.md5
          parameters:
            endpoint: https://eu-central-1.linodeobjects.com
          # or get the model directly from NGC API
          # peoplenet v2.0
          # url: "https://api.ngc.nvidia.com/v2/models/nvidia/tao/peoplenet/versions/pruned_v2.0/zip"

        # model file name, without location
        model_file: resnet34_peoplenet_pruned.etlt  # v2.0 Accuracy: 84.3 Size 20.9 MB

        # configuration of input data and custom preprocessing methods
        input:
          # model input layer name
          layer_name: input_1
          # model input layer shape
          shape: [3, 544, 960]
          # pixel scaling/normalization factor
          scale_factor: 0.0039215697906911373

        # configuration of model output
        output:
          # model output layer names
          layer_names: [output_bbox/BiasAdd, output_cov/Sigmoid]
          # number of detected classes for detector model
          num_detected_classes: 3
          # specify which detected objects are included in output
          objects:
            # object class id
            - class_id: 0
              # label assigned to objects of this class id
              label: person
              selector:
                kwargs:
                  # minimal width of the objects of this class to be included in output
                  min_width: 32
                  # minimal height of the objects of this class to be included in output
                  min_height: 32