MapReader_railspace_London_imago_mundi_2025

MapReader Outputs

Railspace patches and text for London maps inferred with MapReader software using https://huggingface.co/Livingwithmachines/mr_resnest101e_finetuned_OS_6inch_2nd_ed_railspace on Ordnance Survey 6-inch-to-1-mile 2nd edition map sheets from the National Library of Scotland. 

How did we create the London polygon.

Our London polygon was defined as a 20 mile radius from a point in central London:

Files:

• 100meter_patch_df.csv - 586,275 patches data
• 100meter_parent_df.csv - 329 maps metadata
• railspace_predictions_patch_df.csv - 586,275 patches classified as either "no" or "railspace", 556,721 "no", 29,554 "railspace"
• post_processed_railspace_predictions_patch_df.csv - 586,275 patches classified as either "no" or "railspace", 556,880 "no", 29,395 "railspace" 
• geo_predictions_deduplicated_point.csv - georeferenced text spotting predictions for all maps with polygons simplified to points. This file only contains point data.
• geo_predictions_deduplicated_centroid.csv - georeferenced text spotting predictions for all maps with polygons simplified to points. This file contains both polygons and point data for text spotting predictions but will load points as geometry by default, you can update this by setting the geometry as the `polygon` column.

Note: new columns have been added in post processed dataframe with updated label + label index "new_predicted_label" and "new_pred". This post-processing was done using MapReader's context-based post-processing tool [here](https://mapreader.readthedocs.io/en/latest/using-mapreader/step-by-step-guide/5-post-process.html#context-post-processing). We used default confidence threshold of 0.7.