njurs / earth-vision

Jakarta Research Earth Observatory - Library

earth-vision

Earth Vision is a python library for solving computer vision tasks specifically for satellite imagery.

Objective

To ease researcher to run ML pipelines for AI or Deep Learning Applications in solving Earth Observation (EO) tasks.

Installation

We recommend Anaconda as Python package management system and using Python 3.9.

pip:

pip install earth-vision
conda install gdal

From source:

python setup.py install
conda install gdal

GDAL is actually a C++ library with python bindings. That means it relies on underlying C++ code and the package must be built/compiled in a certain manner to be usable with Python. So, we prefer to install it from Anaconda.

Example

from torch.utils.data import DataLoader
from torchvision.transforms import ToTensor, Compose, Normalize
from earthvision.datasets import RESISC45
from earthvision.models.resisc45 import regnet_y_400mf

# Transformation
preprocess = Compose([ToTensor(), 
                      Normalize(mean=[0.3680, 0.3810, 0.3436], 
                                std=[0.1454, 0.1356, 0.1320])])

# Dataset and Dataloader
dataset = RESISC45(root='../dataset', transform=preprocess, download=True)
dataloader = DataLoader(dataset, batch_size=32, shuffle=True)

# Model
model = regnet_y_400mf(pretrained=True)

Features Plans

Feel free to suggest features you would like to see by opening an issue.

• GPU memory optimization [TBD]
• High-level pipeline to integrate varied data sources [TBD]