README.md

---
title: Maya4
emoji: 📊
colorFrom: indigo
colorTo: indigo
sdk: static
pinned: true
license: apache-2.0
thumbnail: >-
  https://cdn-uploads.huggingface.co/production/uploads/68a43a86eea45496edb28ba6/hJyITwdmrofMtEtX8qFmW.png
---

<p align="center">
  <img src="Maya4.png" alt="Maya4 Logo" width="780">
</p>

<h1 align="center">Maya4</h1>

<p align="center">
  Multi-level intermediate SAR representations from Sentinel-1 Stripmap acquisitions
</p>

<p align="center">
  <strong>Level-0 to Level-1</strong> · <strong>Zarr-native</strong> · <strong>Cloud-accessible</strong> · <strong>2 TB</strong>
</p>

---

## Overview

**Maya4** is a curated SAR data resource designed to expose the full progression of **Sentinel-1 Stripmap acquisitions** from **raw radar echoes** to **fully focused Level-1 imagery**.

Unlike conventional datasets that provide only final products, Maya4 preserves and organizes the **intermediate signal representations** generated across the SAR processing chain. This makes the dataset particularly suitable for:

- SAR signal processing research
- physics-aware machine learning
- self-supervised pre-training
- representation learning across processing levels
- algorithm benchmarking and reconstruction studies

The name *Maya4* draws from the concept of **Māyā**: the idea that reality is revealed through successive layers. In the same way, SAR imagery emerges through a sequence of transformations from raw measurements to interpretable image products.

---

## Why Maya4

<table>
  <tr>
    <td valign="top" width="33%">
      <h3>Multi-level access</h3>
      <p>Provides consistent access to multiple SAR processing stages rather than only the final image product.</p>
    </td>
    <td valign="top" width="33%">
      <h3>Research-oriented structure</h3>
      <p>Designed for analysis of information flow, model pre-training, and development of custom SAR pipelines.</p>
    </td>
    <td valign="top" width="33%">
      <h3>Cloud-native delivery</h3>
      <p>Distributed in <strong>Zarr</strong> format for scalable storage, streaming, and computation.</p>
    </td>
  </tr>
</table>

---

## Dataset Access

| Dataset | Access | Mission / Mode | Format | Size |
|---------|--------|----------------|--------|------|
| **Maya4** | [Open bucket](https://huggingface.co/buckets/ESA-philab/Maya4) | Sentinel-1 Stripmap | Zarr | 2 TB |

---

## Processing Chain

A defining feature of Maya4 is its **sharded multi-level organization**, which preserves the major intermediate states of the SAR focusing pipeline.

<p align="center">
  <img src="https://i.ibb.co/Wv7SXd4N/intermediates.jpg" alt="Maya4 intermediate SAR representations" width="100%">
</p>

| Processing Level | Abbrev. | Description | Technical Value |
|------------------|---------|-------------|-----------------|
| **Raw** | `raw` | Unprocessed radar echoes as acquired by Sentinel-1 | Enables custom end-to-end SAR processing and low-level signal analysis |
| **Range Compressed** | `rc` | Echoes compressed in the range dimension using matched filtering | Improves signal-to-noise ratio and resolves scatterers along range |
| **Range Cell Migration Corrected** | `rcmc` | Echoes after compensation of range migration effects | Preserves geometric consistency and prepares the signal for azimuth focusing |
| **Azimuth Compressed** | `ac` | Fully focused SAR image in slant-range geometry | Corresponds to the interpretable focused SAR image product |

---

## Technical Positioning

Maya4 is intended to support work at the intersection of:

- SAR signal processing
- remote sensing foundation models
- self-supervised and masked modeling approaches
- physics-guided representation learning
- inverse problems and reconstruction
- benchmarking of processing-aware architectures

Because the dataset exposes multiple internal stages of SAR formation, it enables experiments that are not possible with image-only repositories.

---

## Key Characteristics

| Attribute | Value |
|-----------|-------|
| **Mission** | Copernicus Sentinel-1 |
| **Acquisition Mode** | Stripmap |
| **Processing Coverage** | Level-0 to Level-1 intermediates |
| **Primary Distribution Format** | Zarr |
| **Access Paradigm** | Cloud-native bucket access |
| **Primary Target Users** | SAR researchers, ML practitioners, remote sensing scientists |

---

## Acknowledgements

Maya4 is based on data from the **Copernicus Sentinel-1 mission** of the **European Space Agency (ESA)**.

Dataset curation and organization are maintained by the **Maya4 organization**.