🚸 Walkthrough on object detection with bounding boxes #49
+1,254
−3
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Context geo-tiles in Python!
Initial draft tutorial on preparing vector bounding box labels for an object detection task. Will be looking at building footprints over Kampong Ayer, Brunei on ESRI World Imagery. Also added intersphinx links for contextily and numpy.
Fsspec-compatible Azure Datalake and Azure Blob Storage access!
Show how to load the Microsoft Building Footprints from GeoParquet files! Using `geopandas.read_parquet` for now (should it become a DataPipe?) to demonstrate things work. Have tried `pyogrio.read_dataframe(asset.href)` but got `DataSourceError: abfs://footprints/global/2022-07-06/ml-buildings.parquet/RegionName=Brunei: No such file or directory`. Might need to figure out https://gdal.org/user/virtual_file_systems.html#vsiadls auth issues somehow.
Show how to turn a numpy.ndarray image into an xarray.DataArray grid with coordinates assigned. Had to first change the image from channel-last to channel-first format (Pytorch/Chainer style) and drop the alpha channel (since there's no mask anyway. Afterwards is just setting the coordinates with xarray/rioxarray.
5 tasks
New GeoPandasRectangleClipper DataPipe to use!
Show how to use small chips (generated with xbatcher) as masks to clip an input vector geometry. Included an important note that GeoPandasRectangleClipper only works with one global vector database, but it will reproject and clip to each chip's crs and bounding box extent. Showed a side by side image and vector polygon visualization, and did a bit of Brunei Tourism marketing 😆 Also had to change `.rio.set_crs` to `.rio.write_crs` to preserve the crs property on xbatcher slicing.
Show how to turn things upside down! First step is to derive geographic bounding boxes from the polygons. Second step is to convert the boxes from geographical coordinates to image/pixel coordinates. Also added an intersphinx link for torchvision to reference the `torchvision.ops.box_convert` function.
Show how to create a list of bounding box tensors, looping over batches of object detection data, and georeferencing of the 'predicted' output. Metadata is critical to tie everything together! Also added an intersphinx link for mmdetection, shoutout to their fantastic Model Zoo!
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Tutorial on preparing vector bounding box labels for an object detection task. Using Microsoft Building Footprints over Kampong Ayer, Brunei 🇧🇳
Preview at https://zen3geo--49.org.readthedocs.build/en/49/object-detection-boxes.html
TODO: