How do you evaluate Chapter 3 Model?

[lenoqt]

As title says, how do you evaluate a CNN? I tried use the same approach on chapter 2 but I can't. I get the following.

`---------------------------------------------------------------------------
RuntimeError Traceback (most recent call last)
in
11 img = img_transforms(img).to(device)
12 cnnet.eval()
---> 13 prediction = F.softmax(cnnet(img), dim=1)
14 prediction = prediction.argmax()
15 cats_pred.append(labels[prediction])

D:\Users\gusta\anaconda3\envs\book-1\lib\site-packages\torch\nn\modules\module.py in _call_impl(self, *input, **kwargs)
725 result = self._slow_forward(*input, **kwargs)
726 else:
--> 727 result = self.forward(*input, **kwargs)
728 for hook in itertools.chain(
729 _global_forward_hooks.values(),

in forward(self, x)
30
31 def forward(self, x):
---> 32 x = self.features(x)
33 x = self.avgpool(x)
34 x = torch.flatten(x, 1)

D:\Users\gusta\anaconda3\envs\book-1\lib\site-packages\torch\nn\modules\module.py in _call_impl(self, *input, **kwargs)
725 result = self._slow_forward(*input, **kwargs)
726 else:
--> 727 result = self.forward(*input, **kwargs)
728 for hook in itertools.chain(
729 _global_forward_hooks.values(),

D:\Users\gusta\anaconda3\envs\book-1\lib\site-packages\torch\nn\modules\container.py in forward(self, input)
115 def forward(self, input):
116 for module in self:
--> 117 input = module(input)
118 return input
119

D:\Users\gusta\anaconda3\envs\book-1\lib\site-packages\torch\nn\modules\module.py in _call_impl(self, *input, **kwargs)
725 result = self._slow_forward(*input, **kwargs)
726 else:
--> 727 result = self.forward(*input, **kwargs)
728 for hook in itertools.chain(
729 _global_forward_hooks.values(),

D:\Users\gusta\anaconda3\envs\book-1\lib\site-packages\torch\nn\modules\conv.py in forward(self, input)
421
422 def forward(self, input: Tensor) -> Tensor:
--> 423 return self._conv_forward(input, self.weight)
424
425 class Conv3d(_ConvNd):

D:\Users\gusta\anaconda3\envs\book-1\lib\site-packages\torch\nn\modules\conv.py in _conv_forward(self, input, weight)
417 weight, self.bias, self.stride,
418 _pair(0), self.dilation, self.groups)
--> 419 return F.conv2d(input, weight, self.bias, self.stride,
420 self.padding, self.dilation, self.groups)
421

RuntimeError: Expected 4-dimensional input for 4-dimensional weight [64, 3, 11, 11], but got 3-dimensional input of size [3, 64, 64] instead
`

[MarcusFra]

This error is due to the wrong tensor size since it represents only a single image.
It needs to be converted to a tensor with a shape of [1, 3, 64, 64] as the model expects a DataLoader object which feeds the model with batches instead of single images. To do that, you can use the torch.unsqueeze() method:

labels = ['cat','fish']

img = Image.open(FILENAME)
img = img_transforms(img).to(device)
img = torch.unsqueeze(img, 0)
cnnnet.eval()
prediction = F.softmax(cnnnet(img), dim=1)
prediction = prediction.argmax()
print(labels[prediction])

in order to add the additional "batch" dimension at index 0.

If you want to predict multiple images you can use a test data loader (like the other data loaders) without that additional line (and like you implemented it before with the list you appended to; maybe you should also use an index or any information to which image your predictions refer to).

I've added the missing line to Chapter 2.ipynb --> #53

[MarcusFra]

Just to quickly add the alternative using a test data data loader.
One solution for named predictions might be the usage of test_data_loader.dataset.samples:

batch_size = 64
test_data_path = "YOURPATH"
test_data = torchvision.datasets.ImageFolder(root=test_data_path,transform=img_transforms, is_valid_file=check_image)
test_data_loader = torch.utils.data.DataLoader(test_data, batch_size=batch_size,shuffle=True)

labels = ['cat', 'fish']
preds = []
# get image names from attribute and convert to list
img_names = list(test_data_loader.dataset.samples)

cnnnet.eval()
for batch in test_data_loader:
    inputs, targets = batch
    inputs = inputs.to(device)
    output = cnnnet(inputs)
    prediction = F.softmax(output, dim=1)
    # get indices with max value; like argmax() but for tensor:
    prediction = torch.max(prediction, dim=1)
    # convert tensor (take only indices, not max values) to list:
    predictions = prediction[1].tolist()
    # convert indices to predicted string cat/fish
    predictions = [labels[ind] for ind in predictions]
    preds.append(predictions)

# convert nested lists to list
preds = [item for sublist in preds for item in sublist]
result = list(zip(preds, img_names))

[lenoqt]

Thanks! Seems to work now.