The images of EMNIST() are flipped and rotated 90 degrees anticlockwise by default

[hyperkai]

🐛 Describe the bug

The images of EMNIST() are flipped and rotated 90 degrees anticlockwise by default as shown below:

from torchvision.datasets import EMNIST

train_data = EMNIST(
    root="data",
    split="byclass",
    train=True,
    download=True,
)

test_data = EMNIST(
    root="data",
    split="byclass",
    train=False,
    download=True
)

import matplotlib.pyplot as plt

def show_images(data):
    plt.figure(figsize=(10, 5))
    col = 5
    for i, (image, label) in enumerate(data, 1):
        plt.subplot(1, col, i)
        plt.title(label)
        plt.axis('off')
        plt.imshow(image)
        if i == col:
            break
    plt.show()

show_images(data=train_data)
show_images(data=test_data)

Versions

import torchvision

torchvision.__version__ # '0.20.1'

[NicolasHug]

Thanks for the report @hyperkai .

From https://www.westernsydney.edu.au/__data/assets/text_file/0019/1204408/EMNIST_Readme.txt I'm reading (emphasis mine):

The dataset is provided in two file formats. Both versions of the dataset contain identical information, and are provided entirely for the sake of convenience. The first dataset is provided in a Matlab format that is accessible through both Matlab and Python (using the scipy.io.loadmat function). The second version of the dataset is provided in the same binary format as the original MNIST dataset as outlined in http://yann.lecun.com/exdb/mnist/

Looking at our code, we are using the latter format, and we're using the exact same code to read EMNIST as for reading MNIST, which would suggest that this orientation was in fact originally intended. Am I missing anything obvious?