README.md

---
license: mit
task_categories:
- image-segmentation
size_categories:
- 100K<n<1M
---

# Dataset Card for NOLDO-S12 Dataset

<!-- Provide a quick summary of the dataset. -->

**NoLDO-S12** is a multi-modal dataset for remote sensing image segmentation from Sentinel-1\&2 images, which contains two splits: 
**SSL4EO-S12@NoL** with <ins>noisy labels</ins> for pretraining, 
and two downstream datasets, **SSL4EO-S12@DW** and **SSL4EO-S12@OSM**, with <ins>exact labels</ins> for transfer learning.

## Dataset Details

<!-- Provide a longer summary of what this dataset is. -->

- **Curated by:** Chenying Liu, Conrad M Albrecht, Yi Wang, Xiao Xiang Zhu
- **License:** MIT
- **Repository:** More details at https://github.com/zhu-xlab/CromSS
- **Paper [arXiv]:** https://arxiv.org/abs/2405.01217
- **Citation:**
  ```Bibtex
  @ARTICLE{liu-cromss,
    author={Liu, Chenying and Albrecht, Conrad M and Wang, Yi and Zhu, Xiao Xiang},
    journal={IEEE Transactions on Geoscience and Remote Sensing}, 
    title={CromSS: Cross-modal pretraining with noisy labels for remote sensing image segmentation}, 
    year={2025},
    volume={},
    number={},
    pages={in press}}
  ```
- **Contents:**
  <table>
  <tr>
    <th>Type</th>
    <th>File</th>
    <th>Description</th>
  </tr>
  <tr>
    <td rowspan="3">Data</td>
    <td>ssl4eo_s12_nol.zip</td>
    <td>SSL4EO-S12@NoL pretraining dataset with noisy labels</td>
  </tr>
  <tr>
    <td>ssl4eo_s12_dw.zip</td>
    <td>SSL4EO-S12@DW downstream dataset with 9-class exact labels from the Google DW project</td>
  </tr>
  <tr>
    <td>ssl4eo_s12_osm.zip</td>
    <td>SSL4EO-S12@OSM downstream dataset with 13-class exact labels from OSM</td>
  </tr>
  <tr>
    <td rowspan="4">weights</td>
    <td>weights-cromss-13B-midFusion-epoch=199.ckpt</td>
    <td>pretrained with CromSS and middle fusion using S1 and 13-band S2</td>
  </tr>
  <tr>
    <td>weights-cromss-13B-lateFusion-epoch=199.ckpt</td>
    <td>pretrained with CromSS and late fusion using S1 and 13-band S2</td>
  </tr>
  <tr>
    <td>weights-cromss-9B-midFusion-epoch=199.ckpt</td>
    <td>pretrained with CromSS and middle fusion using S1 and 9-band S2</td>
  </tr>
  <tr>
    <td>weights-cromss-9B-lateFusion-epoch=199.ckpt</td>
    <td>pretrained with CromSS and late fusion using S1 and 9-band S2</td>
  </tr>
</table>
----------------------------------------------------------------------------------------------------------------

## • SSL4EO-S12@NoL

**SSL4EO-S12@NoL** paired the large-scale, multi-modal, and multi-temporal self-supervised <a href='https://github.com/zhu-xlab/SSL4EO-S12' target='_blank'>SSL4EO-S12</a> dataset with the 9-class noisy labels (NoL) sourced from the Google Dynamic World (DW) project on Google Earth Engine (GEE). To keep the dataset's multi-temporal characteristics, we only retain the S1-S2-noisy label triples from the locations where all 4 timestamps of S1-S2 pairs have corresponding DW labels, resulting in about 41\% (103,793 out of the 251,079 locations) noisily labeled data of the SSL4EO-S12 dataset. SSL4EO-S12@NoL well reflects real-world use cases where noisy labels remain more difficult to obtain than bare S1-S2 image pairs.  

The `ssl4eo_s12_nol.zip` contains the 103,793 DW noisy mask quadruples paired for the <a href='https://github.com/zhu-xlab/SSL4EO-S12' target='_blank'>SSL4EO-S12</a> dataset. The paired location IDs are recorded in `dw_complete_ids.csv`.

----------------------------------------------------------------------------------------------------------------

### • SSL4EO-S12@DW \& SSL4EO-S12@OSM

**SSL4EO-S12@DW** and **SSL4EO-S12@OSM** were constructed for RS image segmentation transfer learning experiments with Sentinel-1/2 data. Both are selected on the DW project’s manually annotated training and validation datasets, yet paired with different label sources from DW and OSM. \
**SSL4EO-S12@DW** was constructed from the DW expert labeled training subset of 4,194 tiles with given dimensions of 510×510 pixels and its hold-out validation set of 409 tiles with given dimensions of 512×512. The human labeling process allows some ambiguous areas left unmarked. We spatial-temporally aligned the S1 and S2 data for the training and test tiles with GEE, leading to 3,574 training tiles and 340 test tiles, that is, a total of 656,758,064 training pixels and 60,398,506 test pixels.\
**SSL4EO-S12@OSM** adopts 13-class fine-grained labels derived from OpenStreetMap (OSM) following the work of <a href='https://osmlanduse.org/#12/8.7/49.4/0/' target='_blank'>Schultz et al.</a> We retrieved 2,996 OSM label masks among the 3,914=3,574+340 DW tiles, with the remaining left without OSM labels. After an automatic check with DW labels as reference assisted by some manual inspection, we construct SSL4EO-S12@OSM with 1,375 training tiles and 400 test tiles, that is, a total of 165,993,707 training pixels and 44,535,192 test pixels.

The `ssl4eo_s12_dw.zip` and `ssl4eo_s12_osm.zip` contain the training and test splits for the two curated downstream datasets. 

The ground-truth mask key for the DW test split is `lulc` (the second layer).

----------------------------------------------------------------------------------------------------------------

## Dataset Card Contact

Chenying Liu (chenying.liu@tum.de; chenying.liu023@gmail.com)