Principle:Kornia Kornia ONNX IO Mapping
| Knowledge Sources | |
|---|---|
| Domains | ONNX, Pipeline_Design |
| Last Updated | 2026-02-09 15:00 GMT |
Overview
Technique of configuring input-output connections between ONNX models in a sequential pipeline.
Description
When chaining multiple ONNX models, the output tensor names of one model must be mapped to the input tensor names of the next. IO mapping specifies these connections as (output_name, input_name) tuples. Without correct mapping, the pipeline cannot route data between models.
This is a configuration pattern that users must define based on knowledge of each model's input/output node names, which can be inspected from the ONNX graph.
Usage
Use when building multi-model ONNX pipelines where automatic name matching is insufficient. Inspect model graphs to determine correct node names.
Theoretical Basis
Given models M1 with outputs {o1, o2} and M2 with inputs {i1, i2}, io_maps defines the routing:
io_maps = [("o1", "i1"), ("o2", "i2")]
The combined graph replaces M1's output nodes with M2's input nodes at the connection points.