Cannot use Inception_v3 with Avalanche

[caesar753]

Hello,

I'm trying to use Avalanche with pretrained CNNs, downloading them from torch.hub

model = torch.hub.load('pytorch/vision:v0.10.0', "inception_v3", pretrained=True, force_reload=True)

it works well with various kind of ResNet (18, 32, 50, ...), but when I use Google Inception_v3 it gives to me this error

TypeError: cross_entropy_loss(): argument 'input' (position 1) must be Tensor, not InceptionOutputs

[HamedHemati]

The output of InceptionV3 is a tuple by default, and Avalanche expects the model's output to be logits. You can disable the auxiliary output in InceptionV3 by setting aux_logits=False when initializing your model:

model = torch.hub.load('pytorch/vision:v0.10.0', "inception_v3", pretrained=True, force_reload=True, aux_logits=False)

[caesar753]

Great, it works!