Multi-Task Face Analysis: Smile, Age, and Gender on GENKI4k and IMDB-Wiki

This repository aggregates facial image data from GENKI4k and IMDB-Wiki for multi-task learning applications. Developed by truongnmt and last updated in 2018, it provides labels for smile detection, age estimation, and gender classification. The collection utilizes MTCNN for face detection and alignment across the combined image sets.

Use Cases

• Training multi-task neural networks to predict smile, age, and gender simultaneously from a single image
• Evaluating MTCNN-based face detection and alignment performance on diverse facial orientations
• Benchmarking age estimation algorithms using the IMDB-Wiki label set

Strengths

• Includes GENKI4k dataset containing 4,000 face images
• Combines three distinct facial attributes (smile, age, gender) for multi-task learning
• Released under the permissive MIT license

Limitations

• IMDB-Wiki labels are known to contain noise due to automated crawling methods
• Dataset has not been updated since 2018
• Exact record count for the combined multi-task subset is not explicitly defined

Provenance

Source

GENKI4k and IMDB-Wiki datasets

Collection Method

Aggregated from existing facial image benchmarks and processed via MTCNN

Time Range

Up to 2018

Freshness

Last updated June 2018; no subsequent updates recorded.