Wrong accuracy reported in Torchvision site?

vision
1

I used a Resnet-18 pretrained on Imagenet as you can see in torchvision library.
The site report 69.7, but I obtained 68.9.
I used the same preprocessing (test set) as reported in torchvision:

  • normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406], 0.229, 0.224, 0.225])
  • torchvision.transforms.Resize(256)
  • torchvision.transforms.CenterCrop(224)

Can anyone help me? Is it possible an error on the table reported in torchvision?
https://pytorch.org/vision/0.8/models.html

2

Did you use model.eval()?

3

Sure.

def test(model, testloader):
    model.eval()
    correct, total = 0, 0
    with torch.no_grad():
        for idx, (data, labels) in enumerate(testloader):
                data, labels = data.cuda(), labels.cuda()
                outputs = model(data)
                _, predicted = torch.max(outputs.data, 1)
                total += labels.size(0)
                correct += (predicted == labels).sum().float().item()

    acc = correct * 100. / total
    err = 100. - acc
    return acc, err
1 Like
4

Can you check this post.

1 Like