• 342×342 (32k)
• 800×800 (131k)
• 2186×2186 (775k)

PIE is provided by the face research group at CMU. a data set of 1166 face images, each of size 32×32, taken in different lighting conditions. The visualization was created with the following steps:

1. Load the PIE data in raw pixel representation. Suppose the data matrix is X, each row for a face image.
2. Calculate 50-Nearest-Neighbor graph, using the Euclidean distances, i.e. A=fastknn(X,50);
3. Supply A to NE using wtsne_p, with over-attraction initialilzation, i.e. Y=wtsne_p(A, true);
4. Place the images according to Y.