Datasets:

buddhi19
/

SyntheticGenV5

pretty_name: SyntheticGenV5
license: mit
task_categories:
  - image-segmentation
task_ids:
  - semantic-segmentation
tags:
  - remote-sensing
  - semantic-segmentation
  - synthetic-data
  - domain-adaptation
  - image
size_categories:
  - 1K<n<10K
configs:
  - config_name: default
    default: true
    drop_labels: true
    data_files:
      - split: train
        path: Train/metadata.csv

SyntheticGenV5

SyntheticGenV5 is a synthetic remote-sensing semantic segmentation dataset (from the paper https://huggingface.co/papers/2602.04749) built for Urban–Rural domain-aware learning.

It keeps the original folder layout and uses Train/metadata.csv to connect each image with its semantic mask and RGB mask.

Why use this dataset?

🌆 Two domains: Urban and Rural
🛰️ Designed for remote-sensing semantic segmentation
🧪 Useful for synthetic augmentation and domain generalization
👀 Includes RGB mask visualizations for easy inspection

Structure

Train/
├── metadata.csv
├── Urban/
│   ├── image_png/
│   ├── mask_png/
│   └── mask_rgb_png/
└── Rural/
    ├── image_png/
    ├── mask_png/
    └── mask_rgb_png/

Metadata Fields

Each row in Train/metadata.csv contains:

image_file_name
mask_file_name
mask_rgb_file_name
domain
source_dataset

Load the dataset

from datasets import load_dataset

ds = load_dataset("buddhi19/SyntheticGenV5")
print(ds["train"][0])

Source

This dataset is derived based on LoveDA

Your downstream segmentation would work way better if you couple this dataset with original LoveDA dataset ;)

LoveDA: A Remote Sensing Land-Cover Dataset for Domain Adaptive Semantic Segmentation

Citation

SyntheticGenV5 / Associated Paper

@misc{wijenayake2026mitigating,
  title={Mitigating Long-Tail Bias via Prompt-Controlled Diffusion Augmentation},
  author={Buddhi Wijenayake and Nichula Wasalathilake and Roshan Godaliyadda and Vijitha Herath and Parakrama Ekanayake and Vishal M. Patel},
  year={2026},
  eprint={2602.04749},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2602.04749}
}

LoveDA

@misc{wang2022lovedaremotesensinglandcover,
      title={LoveDA: A Remote Sensing Land-Cover Dataset for Domain Adaptive Semantic Segmentation}, 
      author={Junjue Wang and Zhuo Zheng and Ailong Ma and Xiaoyan Lu and Yanfei Zhong},
      year={2022},
      eprint={2110.08733},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2110.08733}, 
}

Notes

The original directory layout is preserved.
Train/metadata.csv is used for cleaner loading on Hugging Face.
RGB masks are included mainly for visualization.
This release currently contains the train split.

Acknowledgement

We thank the LoveDA authors for the original benchmark that inspired and supported this dataset.