ESmike/true_imgs_modalities

Dataset Summary

ESmike/true_imgs_modalities is a multimodal dataset containing 4,149 real images at a uniform resolution of 512 × 512, along with multiple derived modalities for each image. This dataset is intended for research and experimentation in computer vision, generative modeling, and multimodal learning.

All images originate from real photographs curated and processed. Each base image is provided alongside additional modality representations (details in the Data Fields section).

If you use this dataset in your research, add the below citation.

Supported Tasks and Use Cases

This dataset can be used for a wide range of tasks, including:

• —Multimodal image-to-image training
• —Generative model conditioning
• —Representation learning
• —Vision–language / multimodal modeling
• —Super-resolution or transformation tasks
• —Benchmarking image encoders

Dataset Structure

Data Fields

• —image: The original real 512×512 image.
• —caption: A short text description of the image. filename: Original filename used during dataset creation. width, height: Original image dimensions before standardization. std512: The main 512×512 standardized real image. colorboost: A color-enhanced version of the image. grayscale: A grayscale transformation of the image. edges: Edge-detected version of the image.

Citation

If you use this dataset, please cite:

@dataset{esmike_true_imgs_modalities,
  author = {Eric Saikali},
  title = {true_imgs_modalities},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/ESmike/true_imgs_modalities}
}

How to Use

from datasets import load_dataset

ds = load_dataset("ESmike/true_imgs_modalities")
example = ds["train"][0]

image = example["image"]
modality = example["std512"]