# IMDB-WIKI

IMDB-WIKI is a large-scale dataset of 523,051 face images with age and gender labels, widely used for training and benchmarking deep learning models in facial age and gender estimation.

IMDB-WIKI is a large-scale dataset for facial age and gender estimation, containing 523,051 face images collected from the Internet Movie Database (IMDB) and Wikipedia. Each image is annotated with a gender label and an age value derived from the associated metadata, such as movie release dates or biographical information. The dataset was introduced in 2015 by researchers Rasmus Rothe, Radu Timofte, and Luc Van Gool at ETH Zurich, and it has become a standard benchmark for evaluating computer vision models in demographic attribute recognition.

The dataset is notable for its size and real-world diversity, as images are drawn from celebrity photos and biographical portraits. However, the age labels are noisy because they are estimated from the difference between the image capture date and the subject's birth date, which can be inaccurate for older photos or when the capture date is unknown. Despite this, IMDB-WIKI has been instrumental in advancing deep learning approaches for age estimation, often serving as a pre-training corpus before fine-tuning on cleaner datasets.

## Construction and Annotation

The IMDB-WIKI dataset was built by automatically crawling images from IMDB and Wikipedia. For IMDB, images were associated with actors and actresses, and the age was computed as the difference between the movie release year and the actor's birth year. For Wikipedia, images were linked to biographical pages, and the age was derived from the page's last modification date or the subject's birth date. Gender was assigned based on the subject's known sex. The final dataset includes 460,723 images from IMDB and 62,328 from Wikipedia, totaling 523,051 images. Faces were detected using a face detector, and each image was cropped to the face region, though some crops contain errors or multiple faces.

## Usage in Deep Learning

IMDB-WIKI is widely used in the [deep-learning](https://www.wikiprompt.org/wiki/deep-learning) community for training and evaluating [neural networks](https://www.wikiprompt.org/wiki/neural-network) on age and gender classification tasks. Many state-of-the-art models, including those based on [residual networks](https://www.wikiprompt.org/wiki/residual-network) and other convolutional architectures, have been pre-trained on IMDB-WIKI. The dataset's large scale helps models learn robust features for facial appearance, but its label noise often requires robust loss functions or data cleaning strategies. Researchers have also used IMDB-WIKI to study the effects of age estimation errors, domain shift, and fairness across demographic groups.

## Benchmarks and Evaluation

IMDB-WIKI is commonly used as a benchmark for age estimation, with metrics such as mean absolute error (MAE) and cumulative score. The dataset is split into training and testing sets, though the official split is not always consistent across studies. Many papers report results on the IMDB-WIKI test set after fine-tuning on cleaner datasets like the Adience or UTKFace. The noisy labels make it a challenging benchmark, and models that achieve low MAE on IMDB-WIKI often generalize well to other age estimation tasks. Gender classification accuracy on IMDB-WIKI is also reported, typically exceeding 95% for modern models.

## Limitations and Ethical Considerations

The dataset has several limitations. The age labels are noisy and biased toward celebrities, who may have professional photos and makeup that differ from the general population. The gender labels are binary and do not account for non-binary identities. Additionally, the dataset was collected without explicit consent from the individuals, raising privacy concerns. Researchers using IMDB-WIKI should be aware of these issues and consider the ethical implications of deploying models trained on it, especially in sensitive applications like surveillance or demographic profiling. The dataset is available for research purposes, but its use in commercial products may require additional scrutiny.

## Related Datasets and Impact

IMDB-WIKI has inspired several follow-up datasets, such as the IMDB-Clean and Wiki-Clean versions that filter noisy labels, and it is often compared with the Adience dataset, which has more accurate age groups but fewer images. The dataset has also been used in studies on [data-augmentation](https://www.wikiprompt.org/wiki/data-augmentation) and [curriculum-learning](https://www.wikiprompt.org/wiki/curriculum-learning) to improve age estimation accuracy. Its impact extends beyond computer vision, as it has been used in [artificial-intelligence](https://www.wikiprompt.org/wiki/artificial-intelligence) research on fairness and bias in facial analysis. The dataset remains a key resource for researchers working on demographic attribute recognition, despite its known limitations.

## See Also

- [machine-learning](https://www.wikiprompt.org/wiki/machine-learning)
- [computer-vision](https://www.wikiprompt.org/wiki/computer-vision)
- facial-recognition

## References

- Rothe, R., Timofte, R., & Van Gool, L. (2015). DEX: Deep EXpectation of apparent age from a single image. In Proceedings of the IEEE International Conference on Computer Vision Workshops.
- Rothe, R., Timofte, R., & Van Gool, L. (2016). Deep expectation of real and apparent age from a single image without facial landmarks. International Journal of Computer Vision.

---
Source: https://www.wikiprompt.org/wiki/imdb-wiki
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-13T03:53:19.538423+00:00