📚 HistVis Dataset

HistVis is a dataset designed to evaluate how text-to-image models represent cultural and historical variations in human activities. It contains images generated by multiple models across temporal prompts and activity categories.

📋 Dataset Structure

The main metadata is stored in dataset.csv, with one row per image. Below is a description of each column:

Column	Description
image_path	Relative path to the image file within the repository. These correspond to generations by different models under specific historical prompts.
model	The name of the text-to-image model used to generate the image (e.g., Flux_Schnell, SD_3, SD_XL).
historical_period	The historical era or century the prompt refers to (e.g., 19th_century, 1920s). This is the temporal condition imposed in the prompt.
universal_human_activity	The prompt used to describe the universal human activity, such as "a person listening to music" or "a person laughing with a friend".
category	The broader conceptual category of the human activity (e.g., Health and Well-being, Art, "Music"). This groups related prompts under common cultural dimensions.

🧾 Prompt Format

Each image in the dataset was generated using the following prompt template:

"a [universal_human_activity] in the [historical_period]"

For example:

• "a person listening to music in the 1950s"
• "a person laughing with a friend in the 19th century"

💻 Using the Dataset

You can access the HistVis dataset using the Hugging Face Datasets library. Below are examples showing how to load and explore the dataset.

Basic Usage

from datasets import load_dataset
import pandas as pd

# Load the dataset metadata (CSV only)
dataset = load_dataset('csv', data_files='https://huggingface.co/datasets/latentcanon/HistVis/resolve/main/dataset.csv')

# Convert to pandas DataFrame for easier manipulation
df = pd.DataFrame(dataset['train'])
print(f"Dataset contains {len(df)} entries")

# View first few entries
print(df.head())