NCCM checksum error about torchgeo HOT 6 CLOSED

I will submit a PR for this issue today.

from torchgeo.

This is annoying. Luckily NCCM is CC-BY-4.0. We should probably rehost the dataset on HF and download it from there instead so the checksum is stable.

from torchgeo.

@shreyakannan1205 @yichiac do we already have a copy of NCCM downloaded somewhere? How big is it? I can upload it to HF (or show you how) and then we can change the download URL/checksum, should be simple.

from torchgeo.

Could also checksum the files inside the archive.

from torchgeo.

I have the downloaded NCCM on my local machine. The size is ~2.4GB.

I agree with @favyen2. The zip has only five files (three TIF masks and two GEE codes). The checksums for all individual files are the same, regardless of the download time.

We can also consider downloading only mask files using separate links instead of the whole zip file.

from torchgeo.

5 files isn't bad, although it does open up the possibility of a zip bomb if we first extract the zip file before checksumming. But we already default to skipping the checksum, and TorchGeo isn't a security-focused application, so I'm not too worried.

I'm fine with either:

1. Download 1 zip file, extract, checksum 3 mask files
2. Download 3 mask files, checksum 3 mask files

Who wants to submit the PR?

from torchgeo.