3 ms·
That's the right idea overall, with a few caveats. 1. Yes but you need to manually inspect and verify that the images are of the right class 5. Images are gra
by e_ameisen 9y ago
That's the right idea overall, with a few caveats.
1. Yes but you need to manually inspect and verify that the images are of the right class
5. Images are grayscale, not only black and white.
Additionally, MNIST and fashion-MNIST have all their objects centered and of similar scale. This is a large part of what makes them a popular first test for any image model: they are very simple to solve as the model need not be very robust to fit the dataset.
- Q6T46nT668w6i3m 9y ago> Additionally, MNIST and fashion-MNIST have all their objects centered and of similar scale. This is a large part of what makes them a popular first test for any image model: they are very simple to solve as the model need not be very robust to fit the dataset. Yeah, this is crucial. Especially when trying to generalize models. It’s easy to verify the usefulness of data augmentation when you can make basic assumptions about the data (e.g. it’s centered).