Inconsistency in inception

🐛 Bug

Hi
I was going through the torchvision code to implement c++ API and I noticed an inconsistency.

In line 298 of inception.py stddev of a class of type BasicConv2d is set to 0.01:

self.conv1.stddev = 0.01

But in line 60 it checks if it's conv2d or linear which a BasicConv2d is none:

if isinstance(m, nn.Conv2d) or isinstance(m, nn.Linear):
   import scipy.stats as stats
   stddev = m.stddev if hasattr(m, 'stddev') else 0.1

I put this code inside the previous if statement:

if hasattr(m, 'stddev'):
    print(m)

and it only printed:

Linear(in_features=768, out_features=1000, bias=True)

so convs inside conv1 BasicConv2d of InceptionAux get initialized with stddev=0.1

is this an error or is it intentional?

Read inception.py around lines 60 and 298, then trace how BasicConv2d and InceptionAux are initialized. Verify whether the differing stddev behavior is intentional and identify the affected convolution path. Done means the behavior is confirmed or corrected consistently, with the result verified against the initialization logic.