Get torch.nn.modules attributes

vision
1

Hi, guys.

So… I’m building a script to convert weights and configuration files from a model. And to do that, I’d like to access the attributes list from the modules. For example:

for name, m in model.named_modules():
t = type(m)
if t is torch.nn.modules.conv.Conv2d:
print(m.in_channels)
print(m.out_channels)
print(m.kernel_size)
print(m.stride)
print(m.dilation)

In that case, above, I defined the attributes to the Conv2d manually. But if the next module is models.common.Conv?

I’d like to know if is it possible to have access to this list?

Tks!

2

vars or dir might be useful in this case:

conv = nn.Conv2d(3, 6, 3, 1, 1)
attrs = vars(conv)
print(', '.join("{}: {}".format(item[0], item[1]) for item in attrs.items()))
print(dir(conv))
3

Tks! It works for me.