from torchvision.models import googlenet, GoogLeNet_Weights

model = googlenet(weights=GoogLeNet_Weights.IMAGENET1K_V1, transform_input=False)

import torch
import torch.nn as nn

x = torch.randn(1, 3, 224, 224, dtype=torch.float)
print(x.shape)

class gnet(nn.Module):
    def __init__(self, model):
        super().__init__()
        self.m = model
        self.softmax = nn.Softmax(-1)

    def forward(self, x):
        return self.softmax(self.m(x))

m = gnet(model)
torch.onnx.export(m, x, "googlenet_2023.onnx", export_params=True, opset_version=13, do_constant_folding=True)

import onnx

model = onnx.load("googlenet_2023.onnx")
inferred_model = onnx.shape_inference.infer_shapes(model)
onnx.save(inferred_model, "googlenet_2023.onnx")

import onnxruntime as ort
import numpy as np
import cv2 as cv

def ort_inference(model_path, x):
    net = ort.InferenceSession(model_path)
    y = net.run([], {"input.1": x})
    return y[0]

def ocv_inference(model_path, x):
    net = cv.dnn.readNet(model_path)
    net.setInput(x)
    y = net.forward()
    return y

x = np.random.randn(1, 3, 224, 224).astype(np.float32)
ort_y = ort_inference("googlenet_2023.onnx", x)
ocv_y = ocv_inference("googlenet_2023.onnx", x)

#print(ort_y)
#print()
#print(ocv_y)

print(ort_y.shape)
print(ocv_y.shape)
print("all_close=", np.allclose(ort_y, ocv_y, atol=1e-05))

import onnxruntime as ort
import numpy as np
import cv2 as cv

def ort_inference(model_path, x):
    net = ort.InferenceSession(model_path)
    y = net.run([], {"input.1": x})
    return y[0]

def ocv_inference(model_path, x):
    net = cv.dnn.readNet(model_path)
    net.setInput(x)
    y = net.forward()
    return y

x = np.random.randn(1, 3, 224, 224).astype(np.float32)
ort_y = ort_inference("googlenet_2023.onnx", x)
ocv_y = ocv_inference("googlenet_2023.onnx", x)

#print(ort_y)
#print()
#print(ocv_y)

print(ort_y.shape)
print(ocv_y.shape)
print("all_close=", np.allclose(ort_y, ocv_y, atol=1e-05))

(base) Air ➜ new cat gen_outs.py
import onnxruntime as ort
import numpy as np
import cv2 as cv

x1 = cv.imread("../../opencv_extra/testdata/dnn/googlenet_0.png")
x1 = cv.dnn.blobFromImage(x1)
x2 = cv.imread("../../opencv_extra/testdata/dnn/googlenet_1.png")
x2 = cv.dnn.blobFromImage(x2)

net = ort.InferenceSession("modern_googlenet.onnx")
y1 = net.run([], {"input.1": x1})[0]
print(y1.shape)
y2 = net.run([], {"input.1": x2})[0]
print(y2.shape)

#y = np.concatenate([y1[np.newaxis, :, :], y2[np.newaxis, :, :]], axis=0)
y = np.concatenate([y1, y2], axis=0)
print(y.shape)
np.save("googlenet_prob_2023.npy", y)