Coarse-Grained 6x6 pixel MNIST Dataset

You can find files of coarse-grained MNIST dataset with 6x6 pixel resolution, and their sample image folders for both test and train data.

Please unzip the “6x6MNIST_Train&Test.zip”, you can find four files of coarse-grained MNIST dataset with 6x6 pixel resolution, and sample image folders for both test and train data.

1. Four files of 6x6 pixel resolution coarse-grained MNIST dataset (test/train for imaging/classification)
I made them from MNIST original digit dataset (https://pjreddie.com/projects/mnist-in-csv/) with 28x28 pixel, by removing outer two pixels from 28x28 to 24x24 and then resolution down conversion from 24x24 to 6x6.

File Name | Matrix Size | How to load the File | Use
1)MNIST_Test_Nox6x6.mat | 10,000x6x6 | load('File Name','U'); | Imaging
2)MNIST_Test_Nox36.mat | 10,000x36 | load('File Name','UU'); | Classification
3)MNIST_Train_Nox6x6.mat | 60,000x6x6 | load('File Name','V'); | Imaging
4)MNIST_Train_Nox36.mat | 60,000x36 | load('File Name','VV'); | Classification

2. Sample images
There are ten sample images of these 6x6 pixel resolution coarse-grained MNIST test and train data in each “SampleImage_MNIST_Test6x6” and “SampleImage_MNIST_Train6x6” folders with file name format of “No.data_label”.
For example, you can find a file of “No1_7” in “SampleImage_MNIST_Test6x6”, it is an image of first (No.1) MNIST 6x6 pixel coarse-grained test data with label of “7”.

It would be grateful if you could demonstrate more than 99 % digit classification, even in such an extremely low 6x6 resolution.

Good luck !