---
title: README
emoji: 🚀
colorFrom: blue
colorTo: green
sdk: static
pinned: false
license: mit
short_description: Official Repository of Pretrained Models on BigEarthNet v2.0
---

# BigEarthNetv2.0 Pretrained Weights
We provide pretrained weights for several different models. 
All models were trained with different seeds.
The weights for the best-performing model (based on Macro Average Precision on the recommended test split) are uploaded.
All models are available as versions using Sentinel-1 only, Sentinel-2 only or Sentinel-1 and Sentinel-2 data.

The order of bands is as follows:  
For Models using Sentinel-1 only: `["VH", "VV"]`  
For Models using Sentinel-2 only: 10m bands, 20m bands = `["B02", "B03", "B04", "B08", "B05", "B06", "B07", "B11", "B12", "B8A"]`  
For Models using Sentinel-1 and Sentinel-2: 10m bands, 20m bands, S1 bands = `["B02", "B03", "B04", "B08", "B05", "B06", "B07", "B11", "B12", "B8A", "VH", "VV"]`

The output classes are in alphabetical order:  
['Agro-forestry areas', 'Arable land', 'Beaches, dunes, sands', 'Broad-leaved forest', 'Coastal wetlands', 
'Complex cultivation patterns', 'Coniferous forest', 'Industrial or commercial units', 'Inland waters', 
'Inland wetlands', 'Land principally occupied by agriculture, with significant areas of natural vegetation', 
'Marine waters', 'Mixed forest', 'Moors, heathland and sclerophyllous vegetation', 
'Natural grassland and sparsely vegetated areas', 'Pastures', 'Permanent crops', 'Transitional woodland, shrub', 
'Urban fabric']


## ConvMixer
[ConvMixer](https://arxiv.org/abs/2201.09792) was trained in the version with a hidden
dimension of 768 and 32 layers (often called ConvMixer-768/32). The weights are available for the following versions:
 - [Sentinel-1 only](https://huggingface.co/BIFOLD-BigEarthNetv2-0/convmixer_768_32-s1-v0.1.1)
 - [Sentinel-2 only](https://huggingface.co/BIFOLD-BigEarthNetv2-0/convmixer_768_32-s2-v0.1.1)
 - [Sentinel-1 and Sentinel-2](https://huggingface.co/BIFOLD-BigEarthNetv2-0/convmixer_768_32-all-v0.1.1)

## ConvNextv2
[ConvNext v2](https://openaccess.thecvf.com/content/CVPR2023/html/Woo_ConvNeXt_V2_Co-Designing_and_Scaling_ConvNets_With_Masked_Autoencoders_CVPR_2023_paper.html) was trained in the "base" configuration. The weights are available for the following versions:
 - [Sentinel-1 only](https://huggingface.co/BIFOLD-BigEarthNetv2-0/convnextv2_base-s1-v0.1.1)
 - [Sentinel-2 only](https://huggingface.co/BIFOLD-BigEarthNetv2-0/convnextv2_base-s2-v0.1.1)
 - [Sentinel-1 and Sentinel-2](https://huggingface.co/BIFOLD-BigEarthNetv2-0/convnextv2_base-all-v0.1.1)

## MLP-Mixer
[MLP-Mixer](https://proceedings.neurips.cc/paper/2021/hash/cba0a4ee5ccd02fda0fe3f9a3e7b89fe-Abstract.html) was trained in the "base" configuration with a patch size of 16. The weights are available for the following versions:
 - [Sentinel-1 only](https://huggingface.co/BIFOLD-BigEarthNetv2-0/mixer_b16_224-s1-v0.1.1)
 - [Sentinel-2 only](https://huggingface.co/BIFOLD-BigEarthNetv2-0/mixer_b16_224-s2-v0.1.1)
 - [Sentinel-1 and Sentinel-2](https://huggingface.co/BIFOLD-BigEarthNetv2-0/mixer_b16_224-all-v0.1.1)

## MobileViT
[MobileViT](https://arxiv.org/abs/2110.02178) was trained in the "small" configuration. The weights are available for the following versions:
 - [Sentinel-1 only](https://huggingface.co/BIFOLD-BigEarthNetv2-0/mobilevit_s-s1-v0.1.1)
 - [Sentinel-2 only](https://huggingface.co/BIFOLD-BigEarthNetv2-0/mobilevit_s-s2-v0.1.1)
 - [Sentinel-1 and Sentinel-2](https://huggingface.co/BIFOLD-BigEarthNetv2-0/mobilevit_s-all-v0.1.1)

## ResNet
[ResNet](https://openaccess.thecvf.com/content_cvpr_2016/html/He_Deep_Residual_Learning_CVPR_2016_paper.html) was trained in the "50" and "101" configurations.
### ResNet-50
The weights for the "50" configuration are available for the following versions:
 - [Sentinel-1 only](https://huggingface.co/BIFOLD-BigEarthNetv2-0/resnet50-s1-v0.1.1)
 - [Sentinel-2 only](https://huggingface.co/BIFOLD-BigEarthNetv2-0/resnet50-s2-v0.1.1)
 - [Sentinel-1 and Sentinel-2](https://huggingface.co/BIFOLD-BigEarthNetv2-0/resnet50-all-v0.1.1)

### ResNet-101
The weights for the "101" configuration are available for the following versions:
 - [Sentinel-1 only](https://huggingface.co/BIFOLD-BigEarthNetv2-0/resnet101-s1-v0.1.1)
 - [Sentinel-2 only](https://huggingface.co/BIFOLD-BigEarthNetv2-0/resnet101-s2-v0.1.1)
 - [Sentinel-1 and Sentinel-2](https://huggingface.co/BIFOLD-BigEarthNetv2-0/resnet101-all-v0.1.1)

### Vision Transformer
[ViT](https://arxiv.org/abs/2010.11929) was trained in the "base" configuration with a patch size of 8. The weights are available for the following versions:
 - [Sentinel-1 only](https://huggingface.co/BIFOLD-BigEarthNetv2-0/vit_base_patch8_224-s1-v0.1.1)
 - [Sentinel-2 only](https://huggingface.co/BIFOLD-BigEarthNetv2-0/vit_base_patch8_224-s2-v0.1.1)
 - [Sentinel-1 and Sentinel-2](https://huggingface.co/BIFOLD-BigEarthNetv2-0/vit_base_patch8_224-all-v0.1.1)