Dongrae/celeb-fbi

Celeb-FBI: Celebrity Full Body Images Dataset

A cleaned and restructured version of the Celeb-FBI dataset containing 7,208 full-body celebrity images with annotations for height, weight, age, and gender.

Dataset Description

This dataset consists of worldwide celebrity images captured in standing, front-facing positions. It is designed for research on human attribute estimation from full-body images, including height, weight, age, and gender prediction tasks.

Dataset Structure

DatasetDict({
    train: Dataset({
        features: ['id', 'image', 'height', 'weight', 'gender', 'age'],
        num_rows: 6487
    })
    test: Dataset({
        features: ['id', 'image', 'height', 'weight', 'gender', 'age'],
        num_rows: 721
    })
})

Features

Statistics

Data Processing

This version of the dataset includes several improvements over the original:

Cleaning steps applied:

• —Converted height from feet to centimeters for standardization
• —Removed implausible values (e.g., heights outside reasonable human range)
• —Missing or invalid values are encoded as -1
• —Fixed typos in original annotations
• —Manual corrections for identified mislabeled samples

Train/test split:

• —Stratified 90/10 split based on height, age, weight buckets, and gender
• —Ensures balanced representation across attribute combinations

Note: Approximately 14% of samples have at least one missing or invalid attribute value (marked as -1). The dataset contains some noise in annotations—users should account for this in their applications.

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("alecccdd/celeb-fbi")

# Access training data
train_data = dataset["train"]

# Example: iterate over samples
for sample in train_data:
    image = sample["image"]
    height = sample["height"]  # in cm, -1 if missing
    weight = sample["weight"]  # in kg, -1 if missing
    gender = sample["gender"]  # 0=male, 1=female
    age = sample["age"]        # -1 if missing
    
# Filter valid samples for a specific attribute
valid_height_samples = train_data.filter(lambda x: x["height"] != -1)

Intended Uses

• —Human attribute estimation research (height, weight, age, gender)
• —Multi-task learning on human body images
• —Benchmarking computer vision models for biometric prediction
• —Study of visual cues for physical attribute estimation

Limitations

• —Images are of celebrities and may not represent the general population
• —Annotation accuracy depends on publicly available biographical data
• —Some noise exists in the annotations; manual corrections were applied where identified but the dataset is not exhaustively verified
• —Limited age range representation at extremes (few samples under 20 or over 80)
• —Height and weight distributions may reflect celebrity demographics

Ethical Considerations

This dataset uses publicly available images of celebrities. Users should be mindful of:

• —Privacy implications when developing attribute estimation systems
• —Potential biases in celebrity image datasets
• —Responsible use in downstream applications

Citation

If you use this dataset, please cite the original paper:

@misc{debnath2024celebfbibenchmarkdatasethuman,
      title={Celeb-FBI: A Benchmark Dataset on Human Full Body Images and Age, Gender, Height and Weight Estimation using Deep Learning Approach}, 
      author={Pronay Debnath and Usafa Akther Rifa and Busra Kamal Rafa and Ali Haider Talukder Akib and Md. Aminur Rahman},
      year={2024},
      eprint={2407.03486},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2407.03486}, 
}

Paper: arXiv:2407.03486