ChangeLing18K

ChangeLing18K is a large-scale dataset of identity-preserving human body transformation pairs spanning three body types: thin, muscle, and fat. It contains 18,573 directed transformation pairs, designed for training and evaluating body-type transfer models.

Dataset Structure

ChangeLing18K/
├── thin/images/         # Images of thin body types
├── muscle/images/       # Images of muscular body types
├── fat/images/          # Images of fat body types
└── transformation_pairs.json

Each image is a 960x1280 JPEG. Filenames follow the format {gender}_{id}_{variant}.jpg (e.g., female_0347_4.jpg), where the variant index distinguishes different images of the same identity captured with varying clothing, backgrounds, etc. The same filename is shared across body-type folders, enabling paired training.

transformation_pairs.json

A dictionary mapping each identity-variant key to a list of valid [source_type, target_type] transformation pairs:

{
  "female_0347_4": [["thin", "fat"], ["muscle", "thin"], ["fat", "thin"], ["fat", "muscle"]],
  "male_0514_9": [["thin", "fat"], ["fat", "muscle"], ["thin", "muscle"], ["muscle", "fat"]],
  ...
}

Pair Statistics

Transformation	Pairs
fat → muscle	3,042
fat → thin	2,989
muscle → fat	3,026
muscle → thin	3,261
thin → fat	2,979
thin → muscle	3,276
Total	18,573

Usage

It is recommended to install the dataset using git-lfs:

git lfs install
git clone https://<your-username>:<your-hf-token>@huggingface.co/datasets/SridharKamath/ChangeLing18K ./ChangeLing18K 

To load a pair, take the key (e.g., female_0347_4), look up the source image at {source_type}/images/female_0347_4.jpg and the target at {target_type}/images/female_0347_4.jpg.

import json
from PIL import Image

pairs = json.load(open("transformation_pairs.json"))

key = "female_0347_4"
for source_type, target_type in pairs[key]:
    source = Image.open(f"{source_type}/images/{key}.jpg")
    target = Image.open(f"{target_type}/images/{key}.jpg")

License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

Citation

If you use this dataset, please cite:

@misc{khandelwal2025ododepthguideddiffusionidentitypreserving,
      title={Odo: Depth-Guided Diffusion for Identity-Preserving Body Reshaping}, 
      author={Siddharth Khandelwal and Sridhar Kamath and Arjun Jain},
      year={2025},
      eprint={2508.13065},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2508.13065}, 
}