README.md

---
license: cc-by-nc-sa-4.0
tags:
- echocardiography
- doppler-imaging
- mlcroissant
configs:
- config_name: default
  data_files:
  - split: metadata
    path: splits.csv
---

# EchoXFlow

[![GitHub](https://img.shields.io/badge/GitHub-EchoXFlow-black?logo=github)](https://github.com/Ahus-AIM/EchoXFlow)

This dataset repository contains Croissant metadata plus one uncompressed tar archive per exam.

## Extraction

Clone or download the dataset repository first. With Git, this creates an `EchoXFlow/` folder:

```bash
git lfs install
git clone https://huggingface.co/datasets/Ahus-AIM/EchoXFlow
cd EchoXFlow
```

The downloaded repository contains `croissant.json` plus one tar archive per exam under `exams/`. Extract every exam
archive into a local `data/` directory to materialize the complete layout referenced by `croissant.json`:

```bash
mkdir -p data
cp croissant.json data/
for archive in exams/*.tar; do
  tar -xf "$archive" -C data
done
export ECHOXFLOW_DATA_ROOT="$PWD/data"
```

After all archives are extracted, paths in `croissant.json` such as `exams/<exam_id>/<recording_id>.zarr` resolve
under `data/`.

## Usage

Use the EchoXFlow code repository for data-reading utilities, task loaders, and export details:

https://github.com/Ahus-AIM/EchoXFlow

## Packaging Notes

The original raw export used many small Zarr chunk files. The per-exam tar layout is used to keep the Hugging Face
repository file count manageable while retaining the original Zarr stores.