/sirfqc

Satellite Image Representations for Quantum Classifiers

Primary LanguagePythonMIT LicenseMIT

Satellite Image Representations for Quantum Classifiers

Code repository for the article "Satellite Image Representations for Quantum Classifiers" in the special issue of Datenbank-Spektrum "Data Management on Quantum Hardware".

Setup

We provide a Dockerfile for an easy setup. Clone the repository and, in the top-level directory, execute:

docker build -t sirfqc .

Then, start the container.

Data

The EuroSAT dataset can be downloaded here. The NWPU-RESISC45 dataset can be downloaded here.

Getting started

To train a model with default parameters (Data: EuroSAT AnnualCrop vs SeaLake, Transformation: VGG16+AE, Classifier: FVQC), simply execute:

python train.py

To get information and help on the parameters and possible arguments for the script, run:

python train.py -h

To train models for a one-versus-rest multiclass classification with default parameters (Data: EuroSAT, Transformation: VGG16+AE, Classifier: FVQC), execute:

python train_ovr.py
Example result for a one-versus-rest multiclass classification of the EuroSAT dataset with VGG16, autoencoder and FVQC.
Confusion matrix as an example result for a one-versus-rest multiclass classification of the EuroSAT dataset with VGG16, autoencoder and FVQC.