Skip to content

Code for the paper "Scene-to-Patch Earth Observation: Multiple Instance Learning for Land Cover Classification".

License

Notifications You must be signed in to change notification settings

JAEarly/MIL-Land-Cover-Classification

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Scene-to-Patch Earth Observation: Multiple Instance Learning for Land Cover Classification

This repo contains the code for the paper Scene-to-Patch Earth Observation: Multiple Instance Learning for Land Cover Classification, published as part of the Tackling Climate Change with Machine Learning Workshop at NeurIPS 2022.

We use Bonfire for the backend MIL functionality.

main_img

Below we break down each of the directories in this repo:

Config

Configuration for model parameters.

Models

Contains the trained model files. Five repeats per configuration.

Out

Interpretability outputs and other figures that are used in the paper

Results

Raw results for our experiments: scene-level RSME and MAE, patch-level mIoU, and pixel-level mIoU.

Scripts

Contains our executable scripts. These are the entry points to our experiments, and should be run from the root of the repo.

Src

Contains our project-specific code. This includes the dataset, high-level model implementations (low-level model code is implemented in Bonfire), and interpretability studies.

About

Code for the paper "Scene-to-Patch Earth Observation: Multiple Instance Learning for Land Cover Classification".

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages