Skip to content

How to deploy onnx models requiring multiple inputs? #45

@breaker-mm

Description

@breaker-mm

Hi there. I was trying to deploy a pointpillars .onnx model using the dnn_inference package. The model takes three tensor lists as input. Normally, I should write an encoder that publishes all three required nitros tensorlist data with different topics. However, I don't know how to connect the encoder with the tensor_rt_node.
image

I read part of the source code of this package and it seems that the tensor_rt_node does not support subscribing to multiple topics. However, in the tensor_rt_inference GXF extension, I found that the model loaded by the TensorRtInference component can fetch data from multiple rx components (inferred from the tick() function) and thus complete the populating of multiple inputs.
I was confused. Do I need to modify the source code for tensor_rt_node?
What should I do?

After further examination of the source code, I have developed some ideas.
Maybe I need to modify the dnn_image_encoder_node.yaml file (adding the DoubleBufferTransmitter components) and modify the nitros::NitrosPublisherSubscriberConfigMap CONFIG_MAP variable accordingly (in tensor_rt_node.cpp file) based on the names of the new components.
Am I thinking right?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions