Add ssTEM dataset

[adamjstewart]

This PR adds a dataset for ISBI Challenge: Segmentation of neuronal structures in EM stacks, which involves imagery from an ssTEM microscope.

Questions:

• I wasn't sure what to call the dataset, as there are several different ISBI challenges per year
• This dataset requires skimage to load, as it includes multi-channel TIF images that PIL does not support. How should we handle this? Should I make the skimage import lazy so that it isn't required to import torchvision?
• This is a segmentation dataset. As such, most transforms like RandomHorizontalFlip and RandomAffine must be applied to both img and target. Is their any precedence for how to handle this? Or should I have a separate transform and target_transform like other datasets?
• The train dataset comes with both data and labels, but the test dataset only comes with data. Users are expected to upload their predictions to the challenge website and have them graded for them. Hopefully this is okay.

[pmeier]

• The train dataset comes with both data and labels, but the test dataset only comes with data. Users are expected to upload their predictions to the challenge website and have them graded for them. Hopefully this is okay.

I'm not sure about this. IMO we should not have a dataset (or in this case a split of the dataset) without labels. @fmassa ?

[codecov-io]

Codecov Report

Merging #900 into master will decrease coverage by 0.44%.
The diff coverage is 25%.

@@            Coverage Diff             @@
##           master     #900      +/-   ##
==========================================
- Coverage    57.9%   57.45%   -0.45%     
==========================================
  Files          46       47       +1     
  Lines        3696     3756      +60     
  Branches      561      572      +11     
==========================================
+ Hits         2140     2158      +18     
- Misses       1429     1473      +44     
+ Partials      127      125       -2

Impacted Files	Coverage Δ
torchvision/datasets/__init__.py	100% <100%> (ø)	⬆️
torchvision/datasets/sstem.py	23.72% <23.72%> (ø)
torchvision/datasets/imagenet.py	21.55% <0%> (ø)	⬆️
torchvision/transforms/transforms.py	82.54% <0%> (+0.64%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 50d54a8...b36b410. Read the comment docs.

[adamjstewart]

@fmassa What's the status of this PR? Is this something that is still wanted, or should I close it?

[adamjstewart]

It seems like this PR has been abandoned by upstream, so I'm going to close it. Feel free to reopen or steal these commits to make a new PR.