Update README.md

README.md

@@ -9,4 +9,10 @@ tags:
 pretty_name: PairedMNIST
 size_categories:
 - 100M<n<1B
----
+---
+
+Simple toy dataset derived from MNIST, by shuffling the dataset to get image pairs. The task is to learn to get the multiplication value of the two figures in the two images.
+
+Each ${split}_data_merged.npy$ file is a (B, 2, 28x28) numpy array, where B is the total number of samples, and 2 is the two images.
+
+Each ${split}_label_merged.npy$ file is a (B, 3) numpy array, where B is the total number of samples, and 3 is the number of labels -- the first one is the figure of the first image, the second one is the figure of the second image, and the third one is not used