Helpful Libraries#
Warning
These libraries should be considered as experimental!
If you find any issues please use the official bug-tracker!
The following is a short list of unofficial BigEarthNet-related libraries:
BigEarthNet Common#
The BigEarthNet Common library provides a collection of high-level tools to better work with the BigEarthNet dataset. Use this library to:
Use any BigEarthNet related constants
Quickly print constants by using a CLI tool
Safely read JSON files
Deterministically multi-hot encode/decode 19/43-class labels
Quickly accessing metadata from a patch for filtering
Country
Season
Original Split
If it is a snowy patch
If it is a cloudy/shadowy patch
Access related S1 or S2 patch
Create CSV sets via a CLI tool
Access example data without having to download the entire BigEarthNet archive
See the BigEarthNet Common documentation for more details.
BigEarthNet GDF Builder#
The BigEarthNet GDF Builder library helps to generate and extend BigEarthNet GeoDataFrame’s. Use this library to:
Convert all the individual JSON files to a common tabular data structure
Easy processing of metadata with GeoDataFrame’s
Assign patches to country
Assign patches to season
Assign patches to original split
Easy processing of the geographical information
Create complex subsets
In-depth statistichal analysis of the meta-data
See the BigEarthNet GDF Builder documentation for more details.
BigEarthNet Patch Interface#
The BigEarthNet Patch Interface is a deep-learning agnostic wrapper for BigEarthNet patches. The main use is to convert the patches to this intermediate format before processing it to a deep-learning specialized format. Internally, the bands as stored as simple NumPy arrays. The interface is very strict during the object creation to ensure that errors in the pipeline are catched early on. It also provides introspection to quickly understand the underlying data structure.
See the BigEarthNet Patch Interface documentation for more details.
BigEarthNet Encoder#
The goal of the BigEarthNet Encoder library is to quickly transform the original BigEarthNet archive into a deep-learning optimized format. The long-term goal is to support multiple output formats. As of now, the only supported target format is the LMDB archive format.
See the BigEarthNet Encoder documentation for more details.