London tree dataset

Summary

This is a dataset of 132 images (1024x1024) of 13944 trees at 33 locations and around the London area.

Functions included (london_trees.ipynb)

• Download
• Tree count
• Convert JSONS to CSVs
• Example image with annotations
• Generate GT points map
• Generate GT gaussian map
• Normalisation (calculate dataset mean and stdev for RGB channels)
• Train test split
• Generate file summary
• Generate lat-lon of each tree

Tree location data

Included is a csv file in the /data folder which contains the latitude and longitude coordinate of every tree.

Other files

• Pytorch dataset configuration files as used in C^3
• Gaussian kernel functions gaussian_functions.py
• Train and test jsons for pytorch dataset files

How to get dataset

Images

To build the images, you will need:

• A GCP account, and a Google Maps Services Static Map API Key (instructions on how to get these available
• in the repo below)
• GMapLoader python package
• Use the 'Download images' section within the notebook to download the images
• Approximate cost is £1 - £1.50 in Google Maps API fees

JSONS

Available here (Google Drive Link)

Ground truth point x,y coordinate .csv files

These are provided in the repo under /gt_points

Ground truth Map .h5 files

Use code 'Generate GT Map' provided in notebook

How to change labels

Images were labeled using labelMe. To alter the labels:

• Create an empty folder
• Move both the image and JSON for the image into this folder (filenames should be the same apart from the extension)
• use python -m labelme to open labelMe
• Use the 'Open Dir' function in label me to point to the directory with the images and JSONS in
• Images should appear in the bottom right window, select the one you want to change, labels will show