Skip to content

Latest commit

 

History

History
106 lines (82 loc) · 4.6 KB

models.md

File metadata and controls

106 lines (82 loc) · 4.6 KB

models section

The models section is a list of defined model settings.

Important

This section only covers the base model config settings (all models will have these settings). Each model framework has its own specific settings that are not covered here. For more information on specific model settings, see the Models section.

Caution

Each model config is a list entry stored as a dictionary.

name

  • name: <string> REQUIRED
  • Default: None

The name key is used to set the model name. The name is used when sending an inference request.

Important

The name key is REQUIRED and must be unique. The name is lower-cased and preserves spaces. YOLO v10 will be lower-cased to yolo v10, TorcH TesT will be lower-cased to torch test.

enabled

  • enabled: <string>
  • yes or no
  • Default: yes

The enabled key is used to enable or disable the model.

description

  • description: <string>
  • Default: None

The description key is used to set the model description. This key and value are not used for anything other than documentation.

type_of

  • type_of: <string> or model_type: <string>
  • object or face or alpr
  • Default: object

The type_of key is used to set the type of model. This is used by the zomi-client to determine how to filter the output from the model.

Note

The type_of key can change what model keys are available! Different combinations of type_of, framework and sub_framework can result in different model keys being available.

framework

  • framework: <string>
  • opencv or trt or ort or torch or coral or http or face_recognition or alpr
  • Default: ort

The framework key is used to set the ML framework to use.

sub_framework

  • sub_framework: <string>
  • See the table below for available sub-frameworks
  • Default: darknet

The sub_framework key is used to set the sub-framework to use.

Important

The sub_framework choices change based on the framework key. This key can be omitted, some models don't process this key.

Available sub-framework values

Framework Sub-framework(s)
opencv darknet, onnx , [caffe, trt, torch, vino, tensorflow] WIP
torch None
ort None
trt None
coral None
http none , rekognition
face_recognition None
alpr openalpr, plate_recognizer, [rekor] WIP

processor

  • processor: <string>
  • cpu or gpu or tpu or none
  • Default: cpu

The processor key is used to set the processor to use for that model.

Tip

When using framework:http, the processor key is ignored/will always be none. When using framework:coral, the processor key is ignored/will always be tpu.

detection_options subsection

The detection_options subsection is where you define the detection settings for the model. Things like confidence thresholds, NMS thresholds, etc.

confidence

  • confidence: <float>
  • Range: 0.01 - 1.0
  • Default: 0.2

The confidence key is used to set the confidence threshold for detection. I recommend 0.2-0.5 to keep the noise down but also allow the client to do some filtering

detect_color

  • detect_color: <string>
  • yes or no`
  • Default: no

The detect_color key is used to ovveride the global color detection enabled flag. color_detection is configured globally.