Adrian Tofting

One process does all jobs by itself

3

First off, I love the ease of use of this project! I am trying to multiprocess the reading of an image (using [rasterio](https://rasterio.readthedocs.io/en/stable/topics/windowed-rw.html). Each row in my dataframe will read...

Add methods for extracting true footprint for sampling valid data only

11

Fix #1330 RasterDatasets may contain nodata regions due to projecting all file to the same CRS, and due to eventual inherit nodata regions in the images. When IntersectionDataset joins this...

Add BalancedRandomGeoSampler balancing positives and negatives

2

### Summary Lets the user balance positives and negatives, sampled from IntersectionDataset of RastersDataset and VectorDataset. ### Rationale May help reduce bias and increase generalisation. The [RandomGeoSampler](https://torchgeo.readthedocs.io/en/latest/api/samplers.html#random-geo-sampler) currently does not...

Support reading from GDAL virtual file systems (e.g. cloud storage)

3

https://github.com/microsoft/torchgeo/blob/9e57f278188ca36348ce8d5c30d5ae2acb19107c/torchgeo/datasets/geo.py#L363-L367 [GDAL virtual file systems](https://gdal.org/user/virtual_file_systems.html) such as reading directly from Google Buckets (`/vsigs/`) are natively supported by rasterio (through gdal). ```python with rasterio.open("/vsigs/my_bucket/.../my_image.tif") as src: src.read() # etc. ``` The...

Allow GeoDataset to list files in VSI path(s)

31

Fix #1398 Fix #1165 This PR remove blockers for reading raster files directly from [Cloud Storage (buckets)](https://gdal.org/user/virtual_file_systems.html) and other GDAL Virtual File Systems.

Stitching: Access patch geo-transform in callback after predict

11

> Now that `train.py` supports both training and inference, I think we should close this PR. I think the best path forward is to modify `predict_step` in our trainers to...

Check if bbox of intersection is valid

4

We experienced that the resulting bounding box of the intersection between hits (files) had `minx > maxx`. This may be because the VectorDatasets consist of points, and the intersection between...

Add methods _dir_exists and _isdir

1

This PR is my best effort to add utility methods for checking if a file or vsi exists, and if it is a directory. (I have no expereince with Pyrex...

Move all dataset open and reproject to single method

8

Fix #3009 Makes it possible for users to customise reprojection logic with minimal overrides. Assumption is that the same reprojection is intended both during `__init__` and `__getitem__`.

This will make testing in #2903 much simpler. Temporarily storing raster bounds as footprint in the metadata file. To compute the actual valid footprint it will need the new method...