Load and store model

ONNX

Load model from ONNX

AIDGE allows loading a DNN model stored in ONNX format. The resulting graph view is created by:

• Converting all ONNX initializers to Producers.

• Converting all ONNX node to nodes with the corresponding Operator, based on the operator’s name. If an operator is unknown, a generic operator is used.

• Adding all the necessary connections between all the nodes of the computational graph.

The feature recquires a dependency to protobuff and at least a backend (in order to store the weights). This feature is available via a Python plugin.