A flexible BigEarthNet encoder that allows one to quickly convert BigEarthNet to a DL-optimization data format.
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.
To simplify the process of working with BigEarthNet, each patch is first converted to a BigEarthNet-Patch-Interface. This interface will guarantee that the data is complete and valid before moving on to creating the desired format. The patch data is internally stored as a NumPy array to keep the data in a framework-agnostic format.
The library should provide all the necessary functionality via a CLI to allow for quick conversion without requiring an in-depth understanding of the library.
As of now, the only supported target format is the LMDB archive format.