Custom Raster Datasets tutorial fails without git repo

[adamjstewart]

Description

If you run the Custom Raster Datasets tutorial on Google Colab, it will fail because it requires test data that is not present after a pip install. We should instead either create fake data with rasterio or download real data from somewhere.

P.S. The plot image doesn't show anything. Is this expected? This tutorial doesn't look very useful as is.

@RitwikGupta

Steps to reproduce

Run the tutorial on Colab. The cell that tries to unzip the test data fails with the following error message:

---------------------------------------------------------------------------
FileNotFoundError                         Traceback (most recent call last)
[<ipython-input-3-304b17c7ea53>](https://localhost:8080/#) in <module>
      2 
      3 data_root = Path("../../tests/data/xview3/")
----> 4 extract_archive(str(data_root / "sample_data.tar.gz"))

3 frames
[/usr/lib/python3.7/gzip.py](https://localhost:8080/#) in __init__(self, filename, mode, compresslevel, fileobj, mtime)
    166             mode += 'b'
    167         if fileobj is None:
--> 168             fileobj = self.myfileobj = builtins.open(filename, mode or 'rb')
    169         if filename is None:
    170             filename = getattr(fileobj, 'name', '')

FileNotFoundError: [Errno 2] No such file or directory: '../../tests/data/xview3/sample_data.tar.gz'

Version

releases/v0.3

[adamjstewart]

Going to replace the xView3 example with a more complex example that downloads real data (maybe Sentinel), overrides both filename_glob and filename_regex, and plots an RGB image.

[RitwikGupta]

@adamjstewart That sounds good to me. This was a contribution with many downfalls, such as displaying an all-white raster (hence why the plot image doesn't show "anything") and relying on a test file that has to be part of the repo itself.

[calebrob6]

@RitwikGupta, do you have a xView3 mini dataset that could be published on Zenodo?

[RitwikGupta]

@calebrob6 there is an xView3 tiny dataset, but there are two issues:

• The tiny dataset is comprised of 9 scenes, but each scene still has two files, VV_dB.tif and VH_dB.tif that are ~1.2 GB each.
• The xView3 T&Cs prevent rehosting the dataset on other services. This is due to a requirement to track access of the data.

[Sandipriz]

Why RasterDataset in colab throws an error?
root = Path('dset-s2')
assert root.exists()
train_imgs = RasterDataset(root=root/'tra_scene', transforms=scale_image, crs='epsg:3395', res=10)

TypeError: RasterDataset.init() got an unexpected keyword argument 'root

[adamjstewart]

@Sandipriz the root argument was renamed to paths in TorchGeo 0.5, see https://torchgeo.readthedocs.io/en/stable/api/datasets.html#torchgeo.datasets.RasterDataset. Sorry for any confusion!