Datasets:

hcarrion
/

ControllableGenDDI

language:
  - en
license: apache-2.0
size_categories:
  - 100K<n<1M
task_categories:
  - image-to-image
  - text-to-image
  - image-classification
pretty_name: cgDDI
dataset_info:
  features:
    - name: image
      dtype: image
    - name: text
      dtype: string
    - name: file_name
      dtype: string
    - name: synthetic_type
      dtype:
        class_label:
          names:
            '0': healthy
            '1': lesion-mapped
            '2': semantic
    - name: prompt_healthy_image_id
      dtype: int32
    - name: prompt_healthy_image_skin_tone
      dtype: int32
    - name: prompt_healthy_image_has_ruler
      dtype: bool
    - name: prompt_disease
      dtype: string
    - name: prompt_disease_malignancy
      dtype: float32
    - name: prompt_text_skin_tone
      dtype: float32
    - name: parametric_synthetic
      dtype: float32
    - name: training_samples
      dtype: float32
  splits:
    - name: train
      num_bytes: 25165990894.36
      num_examples: 266136
  download_size: 9154922115
  dataset_size: 25165990894.36
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
tags:
  - medical

Dataset Card for cgDDI

This repository is the official dataset of Controllable Generation of Diverse Dermatological Imagery for Fair and Efficient Malignancy Classification.

Code Repository: GitHub
Croissant File: croissant_rai.json

🥐 Croissant File 🥐

We host a dedicated and manually extended croissant file, expanding on the one automatically generated by HuggingFace. This file includes RAI fields and additional metadata useful to the research community. It is available on this repo, direct link here: https://huggingface.co/datasets/hcarrion/ControllableGenDDI/resolve/main/croissant_rai.json

Dataset Details

We introduce cgDDI (Controllable Generation of Diverse Dermatological Imagery), a hybrid framework to synthesize realistic healthy skin samples, map rare lesions, and support fair classification.

The dataset contains 266,136 skin-tone-balanced synthetic images across 65+ disease classes:

Healthy synthetics — 309 pixel-perfect in-distribution healthy skin images
Lesion-mapped synthetics — single-sample rare-disease augmentations
Semantic synthetics — textual-inversion + LoRA latent-diffusion samples
Fairness labels — Fitzpatrick skin-tone tags for stratified evaluation

Note: We do not re-host previous artifacts, thus remember to fetch the original DDI dataset at https://ddi-dataset.github.io/ and the sDDI masks from https://github.com/hectorcarrion/FEDD.

Download / Sample Usage

You can download the dataset via the Hugging Face CLI:

# Download dataset via the HF CLI
hf download hcarrion/ControllableGenDDI --repo-type dataset --local-dir data/cgddi

Models

We openly release 65 disease-conditioned LoRA checkpoints, one per condition, on Hugging Face:

Usage Terms

We release this data, code and models with the intent of academic use and to promote fairness research. We do not allow un-ethical usage of these artifacts.

Citation

@inproceedings{carrion2026cgddi,
  title     = {Controllable Generation of Diverse Dermatological Imagery for Fair and Efficient Malignancy Classification},
  author    = {Carri{\\'o}n, H{\\'e}ctor and Norouzi, Narges},
  booktitle = {Medical Image Computing and Computer-Assisted Intervention (MICCAI)},
  year      = {2026},
  publisher = {Springer},
  series    = {Lecture Notes in Computer Science}
}