docs: Improve predict_tile documentation with examples and tips - #1294
docs: Improve predict_tile documentation with examples and tips#1294Prathamesh8989 wants to merge 3 commits into
Conversation
henrykironde
left a comment
henrykironde
left a comment
There was a problem hiding this comment.
Thank you for the contribution @Prathamesh8989
| # Predict on large geospatial tiles using overlapping windows | ||
| # Initialize the DeepForest model | ||
| model = main.deepforest() | ||
| model.use_release() # Load a pretrained tree detection model |
There was a problem hiding this comment.
Use model.load_model(model_name="weecology/deepforest-tree", revision="main")
|
Hi @henrykironde , Thanks for the suggestion! |
jveitchmichaelis
left a comment
jveitchmichaelis
left a comment
There was a problem hiding this comment.
I'm concerned about some apparent hallucination in the code - for example "saving" a dataframe as a TIF, and incorrect assumptions about return types of some methods.
Thanks for the contribution, but please check that the snippets you've posted work.
| ) | ||
|
|
||
| # Visualize predicted bounding boxes (if using in-memory predictions) | ||
| # Note: plot_predictions expects a list of bounding boxes, not a raster file |
There was a problem hiding this comment.
predict_file and predict_tile both return a dataframe. This comment is incorrect.
There was a problem hiding this comment.
Please remove this edit
|
|
||
| # Visualize predicted bounding boxes (if using in-memory predictions) | ||
| # Note: plot_predictions expects a list of bounding boxes, not a raster file | ||
| # predictions = model.predict_image(path) # Optional: for small image testing |
There was a problem hiding this comment.
Remove the "optional" part here, it's potentially confusing to users deciding which one to go with.
| # plot_predictions(predictions) | ||
|
|
||
| # To save the merged raster predictions: | ||
| # predicted_raster.save("predicted_output.tif") |
There was a problem hiding this comment.
This is incorrect, you can't save a dataframe as a TIF.
|
Thanks for the feedback — I’ve updated the documentation to correct the return types, removed the invalid raster-saving example, clarified the visualization section, and reverted the unintended .gitignore change. Looking forward to your review. |
bw4sz
added
the
Awaiting author contribution
|
This pull request has been marked stale because it has been labeled "Awaiting author contribution" with no activity for 30 days. It will be closed in 14 days if no further activity occurs. Comment on the pull request to keep it open. |
4 participants
Summary
This PR enhances the
predict_tileuser guide by adding detailed tips and parameter explanations to improve clarity for new users.Improvements Included
Added Key Parameters
Included a clear breakdown of:
patch_sizepatch_overlapdataloader_strategyNew Example Usage
Added a clean, standalone code block demonstrating tiled prediction on large rasters.
Grammar & Flow
Refined the documentation text for improved readability and a more professional tone.
Code Consistency
modelpandasandosimports in thepredict_filesectionRelated Issue(s)
This change improves guidance for handling large geospatial raster images, particularly around memory management and patch sizing during prediction.
AI-Assisted Development
AI Tools Used