ermongroup / tile2vec

Implementation and examples for Tile2Vec
https://arxiv.org/abs/1805.02855
MIT License
112 stars 40 forks source link

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 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 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".