How to Resize Images in PyTorch: Simple Guide with Examples

To resize images in PyTorch, use torchvision.transforms.Resize which changes the image size to the specified dimensions. Apply it as part of a transform pipeline on PIL images or tensors before feeding them to your model.

Syntax

The main function to resize images in PyTorch is torchvision.transforms.Resize. You specify the new size as either an integer or a tuple of (height, width). This transform can be applied to PIL images or tensors.

• Resize(size): size can be an int or tuple. If int, smaller edge is matched to size keeping aspect ratio.
• Use inside transforms.Compose to chain multiple transforms.

python

from torchvision import transforms

resize_transform = transforms.Resize((128, 128))  # Resize to 128x128 pixels

Example

This example loads an image, resizes it to 128x128 pixels, and converts it to a tensor for model input.

python

from PIL import Image
from torchvision import transforms

# Load an image from file
image = Image.open('sample.jpg')

# Define transform pipeline: resize and convert to tensor
transform = transforms.Compose([
    transforms.Resize((128, 128)),
    transforms.ToTensor()
])

# Apply transform
resized_tensor = transform(image)

# Print shape: should be [channels, height, width]
print(resized_tensor.shape)

Output

torch.Size([3, 128, 128])

Common Pitfalls

Common mistakes when resizing images in PyTorch include:

• Passing a single int to Resize expecting fixed square size, but it resizes smaller edge only, keeping aspect ratio.
• Trying to resize tensors directly without converting to PIL images or using appropriate tensor transforms.
• Not chaining Resize with ToTensor, causing errors when feeding to models.

python

from torchvision import transforms

# Wrong: single int resizes smaller edge, not both dimensions fixed
resize_wrong = transforms.Resize(128)  # Resizes smaller edge to 128, aspect ratio kept

# Right: use tuple for fixed size
resize_right = transforms.Resize((128, 128))

Quick Reference

Function	Description	Example Usage
transforms.Resize	Resize image to given size	transforms.Resize((128, 128))
transforms.ToTensor	Convert PIL image to tensor	transforms.ToTensor()
transforms.Compose	Chain multiple transforms	transforms.Compose([transforms.Resize((128, 128)), transforms.ToTensor()])

Key Takeaways

Use torchvision.transforms.Resize with a tuple to set exact image size.

Resize works on PIL images; convert tensors properly before resizing.

Chain Resize with ToTensor to prepare images for models.

Passing a single int to Resize keeps aspect ratio by resizing smaller edge only.

Always check output tensor shape to confirm resizing worked as expected.