Classifier (model chaining)

Classifiers run on objects detected by a parent model. They must reference a previously defined model using operate_on.

The SDK supports the softmax classifier, which works with any ONNX model that produces a softmax probability distribution. Compatible backbones include:

ResNet (10/18/34/50/101/152)
VGG (16/19)
GoogLeNet / Inception
MobileNet (v1/v2/v3)
EfficientNet (b0–b7)
SqueezeNet, DenseNet, ShuffleNet
DarkNet (19/53)
Vision Transformers (ViT, DeiT, Swin)

models:
  - name: "vehicle-detector"
    type: "detector"
    architecture: "yolo11"
    onnx_file: /path/to/vehicle_model.onnx
    label_file: /path/to/vehicle_labels.txt

  - name: "color-classifier"
    type: "classifier"
    onnx_file: /path/to/color_model.onnx
    label_file: /path/to/color_labels.txt
    operate_on: "vehicle-detector"
    operate_on_classes: [0, 2]

Fields

Field	Type	Required	Default	Description
`architecture`	string	No	`"softmax"`	Classifier architecture (defaults to softmax)
`operate_on`	string	Yes	—	Name of the parent model. Must appear earlier in the list.
`operate_on_classes`	list of int	No	all parent classes	Only classify objects of these class IDs from the parent

Plus the common model fields.

Fields​

Fields