Loading InceptionV3 taking a lot of time

I am trying to load inceptionV3 from torchvision models using the following command:

import torchvision.models as models
inception=models.inception_v3(pretrained=True,progress=True)
But it is taking close to three minutes to load. I am running it on a local server so I dont think computation power is the issue.

Is there anyway to load the model quickly?

1 Like

Are you seeing this long duration only when initializing the model the first time (the checkpoint might be downloaded in the background) or also when you just rerun the same code?

Hello, Every time I am running the code it is happening. The model is already downloaded.

Is your home folder on a local disk or mounted as a network drive?
In the latter case, you could try to set TORCH_HOME to a local folder.

I did that. It doesn’t help. One more thing the other models (e.g. Googlenet, Alexnet) are loading fine.

Which PyTorch and torchvision versions are you using?

torch 1.1.0
torchvision 0.3.0
CUDA 10.0

Any update on this issue? I also met the same issue every time when I load inception v3. My setting is torch: 1.4.0 and torch vision 0.5.0.

1 Like

The slowdown might be related to this issue. Lets continue the discussion in the other thread please.

1 Like