The QMNIST large-scale handwritten digit dataset is reconstructed from the original data found in the NIST Special Database 19 and reconstructed from the previous MNIST DataThe missing 50,000 test images in the test set form the complete QMNIST dataset.

After the reconstruction algorithm, the best match of the distance and the confidence interval measurement, the reconstruction results of MNIST are as follows:

• Two training datasets are enabled - MNIST training set (60,000) and QMNIST training set (60,000)
• Three test datasets are enabled - MNIST test set (10,000), QMNIST10K training set (10,000) and QMNIST50K training set (60,000)

This dataset was released by Cornell University.