You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi Folks,
I've been using torchgeo for loading data and it's working really well. There seems to be a bug in the samplers (verified it in both RandomGeoSampler and GridGeoSampler). When creating a sampler with an RoI, the sampler currently only uses the RoI to select among the available files in the rtree index. If one of the files goes beyond the RoI, that region will be sampled as well. This is specially problematic in datasets composed of large tiffs (like CDL).
@robertomest, thanks for opening an issue! Please let us know if you run into any others or want to contribute in another way. I'll ping you here when #144 is merged.
@robertomest#144 is now ready if you want to test it out. You can clone the repo, check out the feature/zipdatasets branch, and run your code from the same directory or add the directory to your PYTHONPATH. Let me know if you notice any bugs!
Hi Folks,
I've been using torchgeo for loading data and it's working really well. There seems to be a bug in the samplers (verified it in both
RandomGeoSampler
andGridGeoSampler
). When creating a sampler with an RoI, the sampler currently only uses the RoI to select among the available files in the rtree index. If one of the files goes beyond the RoI, that region will be sampled as well. This is specially problematic in datasets composed of large tiffs (like CDL).Example:
I think the problem would be fixed by computing the sampling bounds as the intersection of the hit bounds and the roi
Let me know if you would like me to open a PR and help out on this.
The text was updated successfully, but these errors were encountered: