Hugging Face's logo

Datasets:
tacofoundation
/
sen2venus

Languages:
English
Tags:
sen2venus
sentinel-2
VENµS
venus
super-resolution
harmonization
License:
Dataset card Files Community
The Dataset Viewer has been disabled on this dataset.

Dataset Image

This dataset follows the TACO specification.


SEN2VENµS: A Dataset for the Training of Sentinel-2 Super-Resolution Algorithms

Description

SEN2VENµS is an open dataset used for super-resolution of Sentinel-2 images by leveraging simultaneous acquisitions with the VENµS satellite. The original dataset includes 10m and 20m cloud-free surface reflectance patches from Sentinel-2, with reference spatially-registered surface reflectance patches at 5m resolution from the VENµS satellite, acquired on the same day. This dataset spans 29 locations, totaling 132,955 patches of 256x256 pixels at 5m resolution.

However, a modified version of the original SEN2VENµS dataset has been created. Several changes have been made to adapt the dataset for use within the TACO framework.

Creators

  • Julien Michel
  • Juan Vinasco-Salinas
  • Jordi Inglada
  • Olivier Hagolle

Original dataset

Modifications in Version 2.0.0:

  • All patches are stored in individual geoTIFF files with proper georeferencing and grouped in zip files per site and per category.
  • The dataset now includes 20-meter resolution SWIR bands B11 and B12 from Sentinel-2. Note: There is no high-resolution reference for these bands since the VENµS sensor does not include SWIR bands.

Files Organization

The dataset is divided into separate sub-datasets in individual zip files for each site. Note that the number of patches and pairs may slightly vary compared to version 1.0.0 due to the previous incorrect count.

Number of Patches and Pairs per Site, with VENµS Viewing Zenith Angle
Site Number of Patches Number of Pairs VENµS Zenith Angle
FR-LQ1 4888 18 1.795402
NARYN 3813 24 5.010906
FGMANAUS 129 4 7.232127
MAD-AMBO 1442 18 14.788115
ARM 15859 39 15.160683
BAMBENW2 9018 34 17.766533
ES-IC3XG 8822 34 18.807686
ANJI 2312 14 19.310494
ATTO 2258 9 22.048651
ESGISB-3 6057 19 23.683871
ESGISB-1 2891 12 24.561609
FR-BIL 7105 30 24.802892
K34-AMAZ 1384 20 24.982675
ESGISB-2 3067 13 26.209776
ALSACE 2653 16 26.877071
LERIDA-1 2281 5 28.524780
ESTUAMAR 911 12 28.871947
SUDOUE-5 2176 20 29.170244
KUDALIAR 7269 20 29.180855
SUDOUE-6 2435 14 29.192055
SUDOUE-4 935 7 29.516127
SUDOUE-3 5363 14 29.998115
SO1 12018 36 30.255978
SUDOUE-2 9700 27 31.295256
ES-LTERA 1701 19 31.971764
FR-LAM 7299 22 32.054056
SO2 738 22 32.218481
BENGA 5857 28 32.587334
JAM2018 2564 18 33.718953
Spectral sensitivity response Map of Sentinel-2 coverage on Theia (orange), available VENµS sites (green) and 29 selected sites (red) for the dataset. Source: Michel et al. (2022)

Taco dataset

The Taco dataset has been modified for use within the TACO framework. Each path and timestamp now contain two images:

  • High-Resolution (HR) images: 8 bands (VENµS does not include SWIR bands).
  • Low-Resolution (LR) images: 10 bands (Sentinel-2 includes SWIR bands B11 and B12).

Several preprocessing steps have been applied to the dataset:

  • Handling of negative pixel values: Some Sentinel-2 images contained negative pixel values, which have been masked to 0.
  • No-data values: All no-data values are now set to 65525 for consistency.
  • Data type conversion: The original dataset was stored in int16, and it has been converted to uint16 for storage and processing.
  • Compression for GeoTIFF files: The dataset uses compression settings to reduce storage size while maintaining data quality:
    • Compression algorithm: zstd
    • Compression level: 13
    • Predictor: 2
    • Interleave mode: band

Each sample consists of two corresponding samples: one from Sentinel-2 and the other from VENµS. You can check the details of each band for these samples as shown in the table below:

idx Band Sentinel-2 VENµS
0 B2 Band 2 - Blue - 10m Band 2 - Blue - 10m
1 B3 Band 3 - Green - 10m Band 3 - Green - 10m
2 B4 Band 4 - Red - 10m Band 4 - Red - 10m
3 B5 Band 5 - Red Edge 1 - 20m Band 5 - Red Edge 1 - 10m
4 B6 Band 6 - Red Edge 2 - 20m Band 6 - Red Edge 2 - 10m
5 B7 Band 7 - Red Edge 3 - 20m Band 7 - Red Edge 3 - 10m
6 B8 Band 8 - Near Infrared - 10m Band 8 - Near Infrared - 10m
7 B8A Band 8A - Red Edge 4 - 20m Band 8A - Red Edge 4 - 10m
8 B11 Band 11 - SWIR 1 - 20m -
9 B12 Band 12 - SWIR 2 - 20m -

🔄 Reproducible Example

Open In Colab

Load this dataset using the tacoreader library.

import tacoreader
import rasterio as rio
import matplotlib.pyplot as plt

dataset = tacoreader.load("tacofoundation:sen2venus")

# Read a sample row
row = dataset.read(1000)
row_id = dataset.iloc[1000]["tortilla:id"]
row_lr = row.iloc[0]
row_hr = row.iloc[1]

# Retrieve the data
lr, hr = row.read(0), row.read(1)
with rio.open(lr) as src_lr, rio.open(hr) as src_hr:
    lr_data = src_lr.read([1, 2, 3]) # Blue, Green, Red of Sentinel-2
    hr_data = src_hr.read([1, 2, 3]) # Blue, Green, Red of Neon

# Display
fig, ax = plt.subplots(1, 2, figsize=(10, 5.5))
ax[0].imshow(lr_data.transpose(1, 2, 0) / 1000)
ax[0].set_title(f'LR_{row_id}')
ax[0].axis('off')
ax[1].imshow(hr_data.transpose(1, 2, 0) / 1000) 
ax[1].set_title(f'HR_{row_id}')
ax[1].axis('off')
plt.tight_layout()
plt.show()
drawing

🛰️ Sensor Information

The sensor related to the dataset: sentinel2msi and venus

🎯 Task

The task associated with this dataset: super-resolution

📂 Original Data Repository

Source location of the raw data:https://zenodo.org/records/14603764

💬 Discussion

Insights or clarifications about the dataset: https://huggingface.co/datasets/tacofoundation/sen2venus/discussions

🔀 Split Strategy

All train.

📚 Scientific Publications

Publications that reference or describe the dataset.

Publication 01

  • DOI: 10.3390/data7070096
  • Summary: Boosted by the progress in deep learning, Single Image Super-Resolution (SISR) has gained a lot of interest in the remote sensing community, who sees it as an opportunity to compensate for satellites’ ever-limited spatial resolution with respect to end users’ needs. This is especially true for Sentinel-2 because of its unique combination of resolution, revisit time, global coverage and free and open data policy. While there has been a great amount of work on network architectures in recent years, deep-learning-based SISR in remote sensing is still limited by the availability of the large training sets it requires. The lack of publicly available large datasets with the required variability in terms of landscapes and seasons pushes researchers to simulate their own datasets by means of downsampling. This may impair the applicability of the trained model on real-world data at the target input resolution. This paper presents SEN2VENµS, an open-data licensed dataset composed of 10 m and 20 m cloud-free surface reflectance patches from Sentinel-2, with their reference spatially registered surface reflectance patches at 5 m resolution acquired on the same day by the VENµS satellite. This dataset covers 29 locations on earth with a total of 132,955 patches of 256 × 256 pixels at 5 m resolution and can be used for the training and comparison of super-resolution algorithms to bring the spatial resolution of 8 of the Sentinel-2 bands up to 5 m.
  • BibTeX Citation:
@article{michel2022sen2venmus,
    title        = {Sen2ven$\mu$s, a dataset for the training of sentinel-2 super-resolution algorithms},
    author       = {Michel, Julien and Vinasco-Salinas, Juan and Inglada, Jordi and Hagolle, Olivier},
    year         = 2022,
    journal      = {Data},
    publisher    = {MDPI},
    volume       = 7,
    number       = 7,
    pages        = 96
}

Publication 02

  • Summary: In remote sensing, Single Image Super-Resolution can be learned from large cross-sensor datasets with matched High Resolution and Low Resolution satellite images, thus avoiding the domain gap issue that occurs when generating the Low Resolution image by degrading the High Resolution one. Yet cross-sensor datasets come with their own challenges, caused by the radiometric and geometric discrepancies that arise from using different sensors and viewing conditions. While those discrepancies can be prominent, their impact has been vastly overlooked in the literature, which often focuses on pursuing more complex models without questioning how they can be trained and fairly evaluated in a cross-sensor setting. This paper intends to fill this gap and provide insight on how to train and evaluate cross-sensor Single-Image Super-Resolution Deep Learning models. First, it investigates standard Image Quality metrics robustness to discrepancies and highlights which ones can actually be trusted in this context. Second, it proposes a complementary set of Frequency Domain Analysis based metrics that are tailored to measure spatial frequency restoration performances. Metrics tailored for measuring radiometric and geometric distortion are also proposed. Third, a robust training and evaluation strategy is proposed, with respect to discrepancies. The effectiveness of the proposed strategy is demonstrated by experiments using two widely used cross-sensor datasets: Sen2Venµs and Worldstrat. Those experiments also showcase how the proposed set of metrics can be used to achieve a fair comparison of different models in a cross-sensor setting.

  • BibTeX Citation:

@unpublished{michel:hal-04723225,
    title        = {{Revisiting remote sensing cross-sensor Single Image Super-Resolution: the overlooked impact of geometric and radiometric distortion}},
    author       = {Michel, Julien and Kalinicheva, Ekaterina and Inglada, Jordi},
    year         = 2025,
    month        = {Jan},
    url          = {https://hal.science/hal-04723225},
    note         = {Submitted to IEEE Transactions on Geoscience and Remote Sensing.This work was partly performed using HPC resources from GENCI-IDRIS (Grant 2023-AD010114835)This work was partly performed using HPC resources from CNES Computing Center.},
    pdf          = {https://hal.science/hal-04723225v2/file/tgrs_michel_double_v2.pdf},
    hal_id       = {hal-04723225},
    hal_version  = {v2}
}

Publication 03

  • DOI: 10.5281/zenodo.14603764

  • Summary: SEN2VENµS is an open dataset for the super-resolution of Sentinel-2 images by leveraging simultaneous acquisitions with the VENµS satellite. The dataset is composed of 10m and 20m cloud-free surface reflectance patches from Sentinel-2, with their reference spatially-registered surface reflectance patches at 5 meters resolution acquired on the same day by the VENµS satellite. This dataset covers 29 locations with a total of 132 955 patches of 256x256 pixels at 5 meters resolution, and can be used for the training of super-resolution algorithms to bring spatial resolution of 8 of the Sentinel-2 bands down to 5 meters.

  • BibTeX Citation:

@dataset{julien_michel_2025_14603764,
    title        = {SEN2VENµS, a dataset for the training of Sentinel-2 super-resolution algorithms},
    author       = {Julien Michel and Juan Vinasco-Salinas and Jordi Inglada and Olivier Hagolle},
    year         = 2025,
    month        = {jan},
    publisher    = {Zenodo},
    doi          = {10.5281/zenodo.14603764},
    url          = {https://doi.org/10.5281/zenodo.14603764},
    version      = {2.0.0}
}

🤝 Data Providers

Organizations or individuals responsible for the dataset.

Name Role URL
CNES / Theia data centre producer https://theia.cnes.fr/
Zenodo host https://zenodo.org/record/14603764/

🧑‍🔬 Curators

Responsible for structuring the dataset in the TACO format.

Name Organization URL
Julio Contreras Image & Signal Processing https://juliocontrerash.github.io/

🌈 Optical Bands

Spectral sensitivity response

Spectral sensitivity response of corresponding spectral bands between Sentinel-2 (top) and VENµS (bottom). Source: Michel et al. (2022)

Downloads last month
469