D-Synth: Synthetic Dermoscopic Dataset with Pixel-Perfect 3D Information

D-Synth is the first synthetic dermoscopic dataset providing pixel-perfect 3D ground truth (metric depth, surface normals, camera intrinsics) for monocular depth estimation in dermatology. It was introduced in DermDepth: Toward Monocular Metric Scale 3D Reconstruction Models for Dermatology (Carrión & Norouzi, MICCAI 2026).

Project Page / GitHub | Paper

Overview

• 3,170 rendered dermoscopic samples at 12–20 mm capture distance, 75° field of view
• Per-sample assets (inside sample_XXXXXX/):
  • image.png — RGB rendering
  • depth.png — pixel-perfect metric depth map
  • meta.json — camera intrinsics and other metadata
  • generation_params.json — full rendering parameters
  • (subset) render_rgb.png, render_depth.png, render_meta.json — additional render variants

Sample Usage

You can download the dataset locally using the Hugging Face CLI:

hf download hcarrion/D-Synth --repo-type dataset --local-dir data/dermdepth_train/dsynth

Or programmatically via Python:

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id='hcarrion/D-Synth', 
    repo_type='dataset', 
    local_dir='data/dermdepth_train/dsynth'
)

Generation pipeline

D-Synth extends S-SYNTH (Kim et al., MICCAI 2024) with:

• Per-pixel metric depth export
• Surface normal map export
• Camera intrinsics export
• Multiple camera angles
• Multiple lesions per scene

It inherits S-SYNTH's anatomically-grounded realism stack:

• Probabilistic lesion growth models
• Layered melanosome / blood / lipid models across epidermis / dermis / hypodermis
• Physics-based light scatter across wavelengths and skin tones

Citation

If you use D-Synth, please cite both DermDepth and the underlying S-SYNTH framework:

@inproceedings{carrion2026dermdepth,
  title     = {DermDepth: Toward Monocular Metric Scale 3D Reconstruction Models for Dermatology},
  author    = {Carri{\'o}n, H{\'e}ctor and Norouzi, Narges},
  booktitle = {Medical Image Computing and Computer-Assisted Intervention (MICCAI)},
  year      = {2026}
}

@inproceedings{kim2024ssynth,
  title     = {S-SYNTH: Knowledge-Based, Synthetic Generation of Skin Images},
  author    = {Kim, Andrea and others},
  booktitle = {Medical Image Computing and Computer-Assisted Intervention (MICCAI)},
  year      = {2024}
}

License

CC BY-NC 4.0 (research / non-commercial use). For other uses, please contact the authors.

Related Resources

• Code & training scripts: https://github.com/hectorcarrion/dermdepth
• Fine-tuned DermDepth checkpoints: https://huggingface.co/hcarrion/DermDepth
• Base depth model: https://huggingface.co/Ruicheng/moge-2-vitl-normal