Onnx graphsurgeon add node op with optional inputs

AI & Data Science Deep Learning (Training & Inference) TensorRT
#1

Hi, i’m looking to use graphsurgeon to add a Resize node to an onnx model
https://github.com/onnx/onnx/blob/master/docs/Operators.md#resize

this op allows for 3 optional inputs, but I only want to use one (‘sizes’, along with the required input ‘x’). How do I go about doing this in onnx graphsurgeon?

#2

Hi,
Request you to share the ONNX model and the script if not shared already so that we can assist you better.
Alongside you can try few things:
https://docs.nvidia.com/deeplearning/tensorrt/quick-start-guide/index.html#onnx-export

  1. validating your model with the below snippet

check_model.py

import sys
import onnx
filename = yourONNXmodel
model = onnx.load(filename)
onnx.checker.check_model(model).
2) Try running your model with trtexec command.
https://github.com/NVIDIA/TensorRT/tree/master/samples/opensource/trtexec
In case you are still facing issue, request you to share the trtexec “”–verbose"" log for further debugging
Thanks!

#4

Creating the model:

    model_path = 'model.onnx'
    shape = (1, 224, 224, 3)
    x0 = gs.Variable(name="x0", dtype=np.float32, shape=shape)
    new_shape = (1, 3, 224, 224)
    Y = gs.Variable(name="Y", dtype=np.float32, shape=new_shape)
   nodes = [
        gs.Node(
            op="Transpose", inputs=[x0], outputs=[Y], attrs={"perm": [0, 3, 1, 2]}
        ),
    ]

    graph = gs.Graph(nodes=nodes, inputs=[x0], outputs=[Y])
    graph.cleanup().toposort()
    model = gs.export_onnx(graph)
    print("checking for errors...")
    check = onnx.checker.check_model(model)
    print(check)
    onnx.save(model, model_path)

Modifying the model attempt:

    graph = gs.import_onnx(onnx.load(model_path))
    first_t = [node for node in graph.nodes if node.op == "Transpose"][0]
    resize_out = gs.Variable("resize_out", dtype=np.float32)
    sizes = np.asarray([1, 3, 512, 512])
    size_tensor = gs.Constant(name="resize_sizes", values=sizes)
    resize = gs.Node(
        op="Resize",
        inputs=[first_t.outputs[0], [], [], size_tensor], #What to do here  <----------
        outputs=[resize_out],
        attrs={},
        name="resize_op",
    )

I’m unsure of what to do for the creation of the gs.Node(op=“Resize”) . Resize takes up to four inputs (3 optional), but I only want to use the first and last ones. If I only give two inputs, then it returns “Node (resize_op) has input size 2 not in range [min=3, max=4].”

#5

I found the solution: it’s using gs.Variable.empty() where in the positions for the unused optional parameters
and you need to set the graph opset to 13 (other values may work, 11 does not)