How can we get values of weights from any layer of the model

I want to get the weight & bias values of any layer of choice, how can I access the same? I could get the Kernel Shape, stride, pad, inputs, output for a particular layer. But how do I get the parameter values of a layer from model.onnx file. I am new to Protobuf.

[TMVector]

Hi @holoLens8, here is an example (I’m using mobilenetv2). You can load the model in Netron to explore visually and find the nodes you want.

You will be able to see the names of the node inputs. Here I have selected the first Conv node, and the weight input W is named mobilenetv20_features_conv0_weight. This code snippet will give you the constant (weight matrix, bias, etc.) as a numpy array:

import onnx, onnx.numpy_helper
model = onnx.load("path/to/model.onnx")
[tensor] = [t for t in model.graph.initializer if t.name == "mobilenetv20_features_conv0_weight"]
w = numpy_helper.to_array(tensor)

You can understand how to interpret the numpy array by looking at the operator documentation – for example the Conv operator says that the weight tensor is of shape:

(M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps.

[houseroad]

For example, in python:

model = onnx.load('model.onnx')
weights = model.graph.initializer