PyTorch image data visualization

how to explore and visualise 3 inputs from the training data

You can visualize your input images using libraries like Matplotlib or OpenCV.

Here’s an example of how you can visualize 3 input images from your training data using PyTorch and Matplotlib:

import matplotlib.pyplot as plt
import torchvision

# function to unnormalize and convert tensor to numpy for visualization
def imshow(inp):
    inp = inp.numpy().transpose((1, 2, 0))
    mean = np.array([0.485, 0.456, 0.406])
    std = np.array([0.229, 0.224, 0.225])
    inp = std * inp + mean
    inp = np.clip(inp, 0, 1)
    plt.imshow(inp)
    plt.show()

# Get a batch of training data
inputs, classes = next(iter(train_loader))

# Make a grid from batch
out = torchvision.utils.make_grid(inputs[:3])

imshow(out)

In this script:

1. imshow is a helper function that unnormalizes an image and converts it from a Tensor to a numpy array so it can be displayed by Matplotlib.

2. inputs, classes = next(iter(train_loader)) gets a batch of training data. inputs is a batch of input images, and classes are the labels for those images.

3. out = torchvision.utils.make_grid(inputs[:3]) takes the first three images from the batch and combines them into a single image that can be displayed.

4. imshow(out) displays the grid of images.

This script assumes that your images have been normalized using the mean [0.485, 0.456, 0.406] and standard deviation [0.229, 0.224, 0.225], and that they are loaded using a DataLoader. Adjust it as necessary to match your setup.

Note that this code will work if your images are color (RGB) images. If your images are grayscale or if they have a different color format, you might need to modify the code accordingly.

• pytorch
• python