For my thesis, I developed some tensor-based algorithms for image classification. I have been using the MNIST dataset as benchmark for quite some time now but I want to test my algorithms on other datasets. Because I need images that are already preprocessed (centered, rotated, ...) there aren't much datasets I can use. For example, the COIL datasets contains images of objects that are rotated which isn't good. I have found one other dataset with faces that I can use but are there any other already preprocessed dataset I can use?
1
-
Because I need images that are already preprocessed (centered, rotated, and the COIL datasets contains images of objects that are rotated which isn't good. You need them rotated but rotated isn't good?? Please edit your question.– user4293Commented Mar 15, 2017 at 16:41
Add a comment
|
1 Answer
The CIFAR-10 dataset consists of 60000 32x32 color images in 10 classes, with 6000 images per class. So, that is a decent pre-processing I feel.
You can train on ImageNet which consists of hundreds and thousands of images belonging to thousands of classes, though not exactly pre-processed in terms of uniform image size. Microsoft Common Objects in Context (COCO) and CIFAR-100 Dataset are some other datasets you can try.
