TU Berlin	RSiM	DIMA	BigEarth	BIFOLD

Convnextv2_base pretained on BigEarthNet v2.0 using Sentinel-1 & Sentinel-2 bands

This model was trained on the BigEarthNet v2.0 (also known as reBEN) dataset using the Sentinel-1 & Sentinel-2 bands. It was trained using the following parameters:

Number of epochs: up to 100
- with early stopping
  - after 5 epochs of no improvement
  - based on validation average precision (macro)
- the weights published in this model card were obtained after 15 training epochs
Batch size: 512
Learning rate: 0.001
Dropout rate: 0.15
Drop Path rate: 0.15
Learning rate scheduler: LinearWarmupCosineAnnealing for 1000 warmup steps
Optimizer: AdamW
Seed: 42

The model was trained using the training script of the official BigEarthNet v2.0 (reBEN) repository. See details in this repository for more information on how to train the model given the parameters above.

The model was evaluated on the test set of the BigEarthNet v2.0 dataset with the following results:

Metric	Value Macro	Value Micro
Average Precision	0.692033	0.857302
F1 Score	0.626945	0.759608
Precision	0.751897	0.796884

Example

Example Input (only RGB bands from Sentinel-2)

Example Output - Labels	Example Output - Scores
Agro-forestry areas Arable land Beaches, dunes, sands ... Urban fabric	0.028380 0.569225 0.148004 ... 0.016203

To use the model, download the codes that defines the model architecture from the official BigEarthNet v2.0 (reBEN) repository and load the model using the code below. Note, that you have to install configilm to use the provided code.

from reben_publication.BigEarthNetv2_0_ImageClassifier import BigEarthNetv2_0_ImageClassifier

model = BigEarthNetv2_0_ImageClassifier.from_pretrained("path_to/huggingface_model_folder")

e.g.

from reben_publication.BigEarthNetv2_0_ImageClassifier import BigEarthNetv2_0_ImageClassifier

model = BigEarthNetv2_0_ImageClassifier.from_pretrained(
  "BIFOLD-BigEarthNetv2-0/BENv2-convnextv2_base-all-v0.1.1")

If you use this model in your research or the provided code, please cite the following papers:

CITATION FOR DATASET PAPER

@article{hackel2024configilm,
  title={ConfigILM: A general purpose configurable library for combining image and language models for visual question answering},
  author={Hackel, Leonard and Clasen, Kai Norman and Demir, Beg{\"u}m},
  journal={SoftwareX},
  volume={26},
  pages={101731},
  year={2024},
  publisher={Elsevier}
}