SEN2SR

A radiometrically and spatially consistent super-resolution framework for Sentinel-2

GitHub: https://github.com/ESAOpenSR/sen2sr 🌐

PyPI: https://pypi.org/project/sen2sr/ 🛠️

Preprint: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5247739 📄

Overview
Installation
From 10m and 20m Sentinel-2 bands to 2.5m
From 10m Sentinel-2 bands to 2.5m
From 20m Sentinel-2 bands to 10m
Predict on large images
Estimate the Local Attention Map

Overview

sen2sr is a Python package designed to enhance the spatial resolution of Sentinel-2 satellite images to 2.5 meters using a set of neural network models.

| Model | Description | Run Link | |--------|-------------|---------| | Run SENSRLite | A lightweight SR model optimized for running fast! | <a target="_blank" href="https://colab.research.google.com/drive/1x65GoI5hOfgX61LhtbATSBm7HySUHSw9?usp=sharing"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/></a> | | Run SENSR | Our lightweight SR model! | <a target="_blank" href="https://colab.research.google.com/drive/1MdhdsPwJyV3f0jUgW_WaVeO3-aqA80OG?usp=sharing"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/></a> | | Run LDSR-S2 & SEN2SR | ESAOpenSR's flagship latent diffusion model | <a target="_blank" href="https://colab.research.google.com/drive/1NJuyswsquOLMFc_AP93P_5QcZnbNhGuB?usp=sharing"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/></a> |

Installation

Install the SEN2SRLite version using pip:

pip install sen2sr mlstac git+https://github.com/ESDS-Leipzig/cubo.git

For using the full version of SEN2SR, which employs the Mamba architecture, install as follows:

⚠️ Warning
mamba-ssm requires a GPU runtime and CUDA version > 12 to function properly.
Installation may take a significant amount of time. Please avoid interrupting the process.

Create a fresh Conda environment:

conda create -n test_env python=3.11
conda activate test_env

Install PyTorch with CUDA support:

pip install torch torchvision --index-url https://download.pytorch.org/whl/cu121

Attempt to install mamba-ssm using --no-build-isolation:

pip install mamba-ssm --no-build-isolation

Install the remaining dependencies:

pip install sen2sr mlstac git+https://github.com/ESDS-Leipzig/cubo.git

Adapted from this state-spaces/mamba issue.

From 10m and 20m Sentinel-2 bands to 2.5m

This example demonstrates the use of the SEN2SRLite model to enhance the spatial resolution of Sentinel-2 imagery. A Sentinel-2 L2A data cube is created over a specified region and time range using the cubo library, including both 10 m and 20 m bands. The pretrained model, downloaded via mlstac, takes a single normalized sample as input and predicts a HR output. The visualization compares the original RGB composite to the super-resolved result.
LDSR-S2 is also available.

import mlstac
import torch
import cubo

# Download the model
mlstac.download(
  file="https://huggingface.co/tacofoundation/sen2sr/resolve/main/SEN2SRLite/main/mlm.json",
  output_dir="model/SEN2SRLite",
)

# Load the model
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = mlstac.load("model/SEN2SRLite").compiled_model(device=device)
model = model.to(device)

# Create a Sentinel-2 L2A data cube for a specific location and date range
da = cubo.create(
    lat=39.49152740347753,
    lon=-0.4308725142800361,
    collection="sentinel-2-l2a",
    bands=["B02", "B03", "B04", "B05", "B06", "B07", "B08", "B8A", "B11", "B12"],
    start_date="2023-01-01",
    end_date="2023-12-31",
    edge_size=128,
    resolution=10
)

# Prepare the data to be used in the model, select just one sample 
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
original_s2_numpy = (da[11].compute().to_numpy() / 10_000).astype("float32")
X = torch.from_numpy(original_s2_numpy).float().to(device)

# Apply model
superX = model(X[None]).squeeze(0)

From 10m Sentinel-2 bands to 2.5m

This example demonstrates the use of the SEN2SRLite NonReference_RGBN_x4 model variant to enhance the spatial resolution of only the 10 m Sentinel-2 bands: red (B04), green (B03), blue (B02), and near-infrared (B08). A Sentinel-2 L2A data cube is created using the cubo library for a specific location and date range. The input is normalized and passed to a pretrained non-reference model optimized for RGB+NIR inputs.

import mlstac
import torch
import cubo

# Download the model
mlstac.download(
  file="https://huggingface.co/tacofoundation/sen2sr/resolve/main/SEN2SRLite/NonReference_RGBN_x4/mlm.json",
  output_dir="model/SEN2SRLite_RGBN",
)

# Create a Sentinel-2 L2A data cube for a specific location and date range
da = cubo.create(
    lat=39.49152740347753,
    lon=-0.4308725142800361,
    collection="sentinel-2-l2a",
    bands=["B04", "B03", "B02", "B08"],
    start_date="2023-01-01",
    end_date="2023-12-31",
    edge_size=128,
    resolution=10
)


# Prepare the data to be used in the model
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
original_s2_numpy = (da[11].compute().to_numpy() / 10_000).astype("float32")
X = torch.from_numpy(original_s2_numpy).float().to(device)
X = torch.nan_to_num(X, nan=0.0, posinf=0.0, neginf=0.0)

# Load the model
model = mlstac.load("model/SEN2SRLite_RGBN").compiled_model(device=device)

# Apply model
superX = model(X[None]).squeeze(0)

From 20m Sentinel-2 bands to 10m

This example demonstrates the use of the SEN2SRLite Reference_RSWIR_x2 model variant to enhance the spatial resolution of the 20 m Sentinel-2 bands: red-edge (B05, B06, B07), shortwave infrared (B11, B12), and near-infrared (B8A) to 10 m.

import mlstac
import torch
import cubo

# Download the model
mlstac.download(
  file="https://huggingface.co/tacofoundation/sen2sr/resolve/main/SEN2SRLite/Reference_RSWIR_x2/mlm.json",
  output_dir="model/SEN2SRLite_Reference_RSWIR_x2",
)

# Create a Sentinel-2 L2A data cube for a specific location and date range
da = cubo.create(
    lat=39.49152740347753,
    lon=-0.4308725142800361,
    collection="sentinel-2-l2a",
    bands=["B02", "B03", "B04", "B05", "B06", "B07", "B08", "B8A", "B11", "B12"],
    start_date="2023-01-01",
    end_date="2023-12-31",
    edge_size=128,
    resolution=10
)

# Prepare the data to be used in the model
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
original_s2_numpy = (da[11].compute().to_numpy() / 10_000).astype("float32")
X = torch.from_numpy(original_s2_numpy).float().to(device)

# Load the model
model = mlstac.load("model/SEN2SRLite_Reference_RSWIR_x2").compiled_model(device=device)
model = model.to(device)

# Apply model
superX = model(X[None]).squeeze(0)

Predict on large images

This example demonstrates the use of SEN2SRLite NonReference_RGBN_x4 for super-resolving large Sentinel-2 RGB+NIR images by chunking the input into smaller overlapping tiles. Although the model is trained to operate on fixed-size 128×128 patches, the sen2sr.predict_large utility automatically segments larger inputs into these tiles, applies the model to each tile independently, and then reconstructs the full image. An overlap margin (e.g., 32 pixels) is introduced between tiles to minimize edge artifacts and ensure continuity across tile boundaries.

import mlstac
import sen2sr
import torch
import cubo

# Create a Sentinel-2 L2A data cube for a specific location and date range
da = cubo.create(
    lat=39.49152740347753,
    lon=-0.4308725142800361,
    collection="sentinel-2-l2a",
    bands=["B02", "B03", "B04", "B05", "B06", "B07", "B08", "B8A", "B11", "B12"],
    start_date="2023-01-01",
    end_date="2023-12-31",
    edge_size=1024,
    resolution=10
)

# Prepare the data to be used in the model
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
original_s2_numpy = (da[11].compute().to_numpy() / 10_000).astype("float32")
X = torch.from_numpy(original_s2_numpy).float().to(device)
X = torch.nan_to_num(X, nan=0.0, posinf=0.0, neginf=0.0)

# Load the model
model = mlstac.load("model/SEN2SRLite").compiled_model(devic

SEN2SR

Install / Use

README

SEN2SR

A radiometrically and spatially consistent super-resolution framework for Sentinel-2

Table of Contents

Overview

Installation

From 10m and 20m Sentinel-2 bands to 2.5m

From 10m Sentinel-2 bands to 2.5m

From 20m Sentinel-2 bands to 10m

Predict on large images