Testing 1 Image VS 1 Dataset

1

Hello all,

I have a working Pytorch ConvNet program. It tests Datasets and runs rather slow, but effectively.

Would anyone know how to test images separately? I want to test 1 image instead of n number of images.

Thank you

2

You can load and process a single image, e.g. via PIL.Image.open and torchvision.transforms.
Once you have created the image tensor for this single input image, you can add a batch dimension via x = x.unsqueeze(0) and pass it to the model.
Let me know, if this works for you.

3

Thank you for the tips, ptrblck. I’m going to test it out right now.

What does x.unsqueeze(0) do? (What would happen if I didn’t use this function call and passed my image to my current model?)

4

unsqueeze adds a dimension at the specified position (here 0 in this example).

1 Like
5

Appreciate the help, harsha_g :smile:.

I’m getting the error: Attribute Error: ‘Compose’ object has no attribute ‘unsqueeze’.

Here’s my code for testing my image on the model, any help would be great!

‘’’

Load and Process a Single Image

new_image = Image.open(‘new_image.jpg’)

transformed_image = transforms.Compose([

transforms.ToPILImage(),

transforms.Resize((28, 28)),

transforms.RandomGrayscale(p = 1.0), # Transforms the image to a gray-scale

transforms.ToTensor(),

])

Add a Batch Dimension

transformed_image = transformed_image.unsqueeze(0)

check_image(transformed_image, model)

6

transformed_image is the transformation, not the actual transformed image.
You would have to define the transformation first and pass the input image to it.
Use this code instead:

new_image = Image.open(‘new_image.jpg’)

transform = transforms.Compose([
    transforms.ToPILImage(),
    transforms.Resize((28, 28)),
    transforms.RandomGrayscale(p = 1.0), # Transforms the image to a gray-scale
    transforms.ToTensor(),
])

transformed_image = transform(new_image)
transformed_image = transformed_image.unsqueeze(0)
check_image(transformed_image, model)
7

This is the error that I am getting as a result of the code:
‘’’

Traceback (most recent call last):

File “/Users/Josh/Desktop/convolutional_neural_networks/pytorch_convolutional_neural_network_mnist.py”, line 233, in
transformed_image = transform(new_image)
File “/Library/Frameworks/Python.framework/Versions/3.8/lib/python3.8/site-packages/torchvision/transforms/transforms.py”, line 61, in call
img = t(img)
File “/Library/Frameworks/Python.framework/Versions/3.8/lib/python3.8/site-packages/torchvision/transforms/transforms.py”, line 127, in call
return F.to_pil_image(pic, self.mode)
File “/Library/Frameworks/Python.framework/Versions/3.8/lib/python3.8/site-packages/torchvision/transforms/functional.py”, line 100, in to_pil_image
raise TypeError(‘pic should be Tensor or ndarray. Got {}.’.format(type(pic)))
TypeError: pic should be Tensor or ndarray. Got <class ‘PIL.JpegImagePlugin.JpegImageFile’>.
‘’’

Am I doing something wrong?

8

Ah sorry, ToPILImage is of course not needed, if you already provide the input as a PIL.Image.
I just copied your transformation without checking it. Just remove transforms.ToPILImage() and rerun the code.

1 Like
9

I see now.

Thanks again for your help, ptrblck!

10

This is so frustrating…

I’m still getting an error, which doesn’t make a whole lot of sense considering the fact I turned the colored image into Grayscale.

Traceback (most recent call last):
  File "/Users/Josh/Desktop/convolutional_neural_networks/pytorch_convolutional_neural_network_mnist.py", line 243, in <module>
    check_image(transformed_image, model)
  File "/Users/Josh/Desktop/convolutional_neural_networks/pytorch_convolutional_neural_network_mnist.py", line 210, in check_image
    scores = model(image)
  File "/Library/Frameworks/Python.framework/Versions/3.8/lib/python3.8/site-packages/torch/nn/modules/module.py", line 550, in __call__
    result = self.forward(*input, **kwargs)
  File "/Users/Josh/Desktop/convolutional_neural_networks/pytorch_convolutional_neural_network_mnist.py", line 55, in forward
    x = activation_functions.relu(self.convolutional_input_layer(x))
  File "/Library/Frameworks/Python.framework/Versions/3.8/lib/python3.8/site-packages/torch/nn/modules/module.py", line 550, in __call__
    result = self.forward(*input, **kwargs)
  File "/Library/Frameworks/Python.framework/Versions/3.8/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 349, in forward
    return self._conv_forward(input, self.weight)
  File "/Library/Frameworks/Python.framework/Versions/3.8/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 345, in _conv_forward
    return F.conv2d(input, weight, self.bias, self.stride,
RuntimeError: Given groups=1, weight of size [8, 1, 3, 3], expected input[1, 3, 28, 28] to have 1 channels, but got 3 channels instead

Any help appreciated!

11

RandomGrayscale will return a 3 channel image as described in the docs:

If input image is 1 channel: grayscale version is 1 channel - If input image is 3 channel: grayscale version is 3 channel with r == g == b

Use transforms.Grayscale(num_output_channels=1), if you want to get a single channel output.

1 Like
12

Great! Your solution is working well.

I have one more thing before this thread can be locked: ‘I got an answer which is tensor([3]). How can I convert this to a string with the classification of the fourth index?’

Thank you for your help.