Add Vaihingen Segmentation

[isaaccorley]

This PR adds the Vaihingen dataset from the WG II/4 2D Semantic Labeling Contest.

Notes:

• I used the specified colormap found here.
• All labels are provided but I split the dataset by filenames into the train/test sets used in the challenge.
• The images are not georefenced but I'm reading them using rasterio and it complains that they aren't georeferenced in the tests.
• Images are all of different sizes
• I was able to abstract the file structure away enough that in a future PR we can make a parent class to combine the Potsdam/Vaihingen datasets or reduce copy/pasted code.

Examples:

[adamjstewart]

The images are not georefenced but I'm reading them using rasterio and it complains that they aren't georeferenced in the tests.

I think that means we just need to read them with PIL instead?

[adamjstewart]

Potsdam and Vaihingen seem very similar, I wonder if they belong in a single isprs.py file or something like that. I guess it depends on how many datasets ISPRS has and whether there is any logical grouping between them.

[isaaccorley]

Refactored to load with PIL. Changed everywhere to Vaihingen2D.

[calebrob6]

RIP tmp.ipynb