Get startedGet started for free

Image dataset

Let's start with building a Torch Dataset of images. You'll use it to explore the data and, later, to feed it into a model.

The training data for the cloud classification task is stored in the following directory structure:

clouds_train
  - cirriform clouds
    - 539cd1c356e9c14749988a12fdf6c515.jpg
    - ...
  - clear sky
  - cumulonimbus clouds
  - cumulus clouds
  - high cumuliform clouds
  - stratiform clouds
  - stratocumulus clouds

There are seven folders inside clouds_train, each representing one cloud type (or a clear sky). Inside each of these folders sit corresponding image files.

The following imports have already been done for you:

from torchvision.datasets import ImageFolder
from torchvision import transforms

This exercise is part of the course

Intermediate Deep Learning with PyTorch

View Course

Exercise instructions

  • Compose two transformations, the first, to parse the image to a tensor, and one to resize the image to 128 by 128, assigning them to train_transforms.
  • Use ImageFolder to define dataset_train, passing it the directory path to the data ("clouds_train") and the transforms defined earlier.

Hands-on interactive exercise

Have a go at this exercise by completing this sample code.

# Compose transformations
train_transforms = ____([
    transforms.____,
    transforms.____,
])

# Create Dataset using ImageFolder
dataset_train = ____(
    ____,
    transform=____,
)
Edit and Run Code