How to preprocess image before using model with onnxruntime?

[XantaKross]

Hey,

I was succesful in trying to run the .onnx model through opencv and python with a input size of (3, 640, 480) which becomes (1, 3, 480, 640) after the

print(img.shape) # (3, 640, 480) blob = self.net.blobFromImage(...) self.net.setInput(blob) print(blob.shape) # (1,3,480,640)

But I wish instead to run the model with onnxruntime inference. Which instead requires a input of size (10, 3, 32, 32)? Am I supposed to

1. Resize the image to 32x32 before as input? And add another dimension?
2. Break the image into 32x32 small patches then stack them into 10 different layers?
3. Do something else?

[michalgrzyska]

Bump for this question, I cannot understand that input size as well. 32x32 makes no sense in case of image object recognition, same as batch = 10

[JoMe2704]

The specs for the input dimension in the ONNX file seem to be wrong. You can fix it as follows:

import onnx
import math

model = onnx.load_model('../models/onnx/centerface.onnx')
d = model.graph.input[0].type.tensor_type.shape.dim
print(d)
d[0].dim_value = 1
d[0].dim_param = 'batch size'
d[2].dim_param = 'height'
d[3].dim_param = 'width'
print(d)
print(model.graph.input[0].type.tensor_type.shape.dim)
for output in model.graph.output:
    d = output.type.tensor_type.shape.dim
    print(d)
    d[0].dim_value = 1
    d[0].dim_param = 'batch size'
    d[2].dim_param = 'h'
    d[3].dim_param = 'w'
    print(d)
onnx.save_model(model,"../models/onnx/centerface_flex.onnx" )