doc/dataset_toothfairy2.md

# Preparing ToothFairy2 Data

**MedVision** is designed to download raw data from source and the annotations from our Huggingface dataset repo. For access to the raw data of the ToothFairy2 dataset, you need to register at https://ditto.ing.unimore.it/toothfairy2/.

This tutorial is about:

- What data to download from the source in order to use the annotations from MedVision?
- How to prepare data?

## What data to download?

- Download and extract data from source

```
# Files from ToothFairy2 source
├── Dataset112_ToothFairy2
	├── imagesTr
		├── ToothFairy2F_001_0000.mha
		├── ...
	├── labelsTr
		├── ToothFairy2F_001.mha
		├── ...
```

## How to prepare data?

⚠️ DO NOT change file or folder names. See why [here](https://huggingface.co/datasets/YongchengYAO/MedVision/blob/main/src/medvision_ds/datasets/ToothFairy2/download.py).

1. Put these folders into `ToothFairy2` and upload the **compressed** file to your HF dataset repo

   ```
   # DO NOT use other folder names
   ├── ToothFairy2
   	├── imagesTr
   	├── labelsTr
   ```

   Your HF repo should contain `ToothFairy2.zip`

2. Set env var whenever you load ToothFairy2 data from MedVision

   ```python
   import os
   os.environ["MedVision_ToothFairy2_HF_ID"] = <your-hf-dataset-id>
   ```