Dataset_comparison

1. Udacity challenge

LINK: https://www.kaggle.com/kumaresanmanickavelu/lyft-udacity-challenge/discussion/101832

12 classes, 5000 images, possibly too much here

2. Ariel imagery

LINK: https://www.kaggle.com/humansintheloop/semantic-segmentation-of-aerial-imagery

6 classes, 72 images Looks good for errors too x4 with rotations?

3. Drone dataset

400 images Questionable privacy LINK: https://www.kaggle.com/bulentsiyah/semantic-drone-dataset

4. Tensorflow cityscape semantic segmentation

LINK: https://www.tensorflow.org/datasets/catalog/cityscapes https://www.cityscapes-dataset.com/

30 classes

5. Multidigit MNIST

LINK https://www.kaggle.com/farhanhubble/multimnistm2nist?select=segmented.npy

5000 images, 11 classes

Good number, but not sure about labelling error potential for different classes