https://github.com/ermongroup/tile2vec

Implementation and examples for Tile2Vec
https://github.com/ermongroup/tile2vec

Last synced: 9 days ago
JSON representation

Implementation and examples for Tile2Vec

• Host: GitHub
• URL: https://github.com/ermongroup/tile2vec
• Owner: ermongroup
• License: mit
• Created: 2018-02-12T08:37:44.000Z (over 6 years ago)
• Default Branch: master
• Last Pushed: 2024-01-25T17:25:10.000Z (9 months ago)
• Last Synced: 2024-07-31T20:49:52.760Z (3 months ago)
• Language: Python
• Homepage: https://arxiv.org/abs/1805.02855
• Size: 8.62 MB
• Stars: 111
• Watchers: 12
• Forks: 39
• Open Issues: 3
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# tile2vec

Implementation and examples for Tile2Vec

## Getting data

Follow the follow instructions for getting the necessary data for the provided examples.

### Getting pre-trained NAIP model

To get a Tile2Vec model that has been pre-trained on the NAIP dataset, download the weights provided [here](https://www.dropbox.com/s/bvzriiqlcof5lol/naip_trained.ckpt?dl=0) and save the file in the models directory: `tile2vec/models/naip_trained.ckpt`.

### Getting tiles

Unzip `data/tiles.zip` in the data directory, i.e., `tile2vec/data/tiles`. Within this folder, you should find 1000 tiles saved as Numpy arrays named `{1-1000}tile.npy` and 1 file named `y.npy` that contains the CDL labels corresponding to these 1000 tiles. 

### Getting triplets

Download tile triplet dataset for the training example [here](https://www.dropbox.com/s/afw3cbvo7sjerru/triplets.zip?dl=0) and then unzip in the data directory, i.e., `tile2vec/data/triplets`. This directory contains 100k tile triplets, with each triplet identified by its index and "anchor", "neighbor", or "distant".