I want use the GoogLeNet model but something wrong

111186 · January 2, 2020, 1:24pm

I have two questions ：
first：what’s the different between googlenet.py and inception.py if I want to use GoogLeNet model which is the suitable.

second：

I want konw if i want to get the output’s size which function is right. when i use the vgg model the shape is ok.
thanks so much

Eta_C · January 2, 2020, 1:52pm

github.com

pytorch/vision/blob/d2c763e14efe57e4bf3ebf916ec243ce8ce3315c/torchvision/models/googlenet.py#L19


from torch import Tensor
from .utils import load_state_dict_from_url


__all__ = ['GoogLeNet', 'googlenet', "GoogLeNetOutputs", "_GoogLeNetOutputs"]


model_urls = {
    # GoogLeNet ported from TensorFlow
    'googlenet': 'https://download.pytorch.org/models/googlenet-1378be20.pth',
}


GoogLeNetOutputs = namedtuple('GoogLeNetOutputs', ['logits', 'aux_logits2', 'aux_logits1'])
GoogLeNetOutputs.__annotations__ = {'logits': Tensor, 'aux_logits2': Optional[Tensor],
                                    'aux_logits1': Optional[Tensor]}


# Script annotations failed with _GoogleNetOutputs = namedtuple ...
# _GoogLeNetOutputs set here for backwards compat
_GoogLeNetOutputs = GoogLeNetOutputs




def googlenet(pretrained=False, progress=True, **kwargs):
    r"""GoogLeNet (Inception v1) model architecture from

As you can see, GoogleNetOutputs is a namedtuple, not a tensor.

Oli · January 2, 2020, 1:52pm

GoogLeNet returns a named tuple and thus behave differently than VGG. See source

GoogLeNetOutputs = namedtuple('GoogLeNetOutputs', ['logits', 'aux_logits2', 'aux_logits1'])

Try

output = net(x)
output_logits = output.logits
print(output_logits.shape)

111186 · January 3, 2020, 1:31am

Thank you for your code it is OK. could you answer my first question .thanks again

111186 · January 3, 2020, 1:33am

Thank you so much:grinning:

fbueller · November 11, 2020, 3:58pm

googlenet.py: GoogLeNet aka Inception v1
inception.py: Inception v3

For more info:

import torchvision
help(torchvision.models.googlenet)
help(torchvision.models.inception_v3)