IMDB-Clean: A Novel Benchmark for Age Estimation in the Wild

Scripts for creating the IMDB-Clean dataset for age estimation and gender classification.

If you use this repository in your research, we kindly rquest you to cite the following paper:

@article{lin2021fpage,
      title={FP-Age: Leveraging Face Parsing Attention for Facial Age Estimation in the Wild}, 
      author={Yiming Lin and Jie Shen and Yujiang Wang and Maja Pantic},
      year={2021},
      eprint={2106.11145},
      journal={arXiv},
      primaryClass={cs.CV}
}

Updates

• 2023-10: Added RetinFace Resnet50 face detector predicions for IMDB-Clean-1024 to csvs/imdb_1024_retinaface_predictions.csv, the number of predictions is smaller than the number of images because some faces are not detected. See visualisation for examples.

Introduction

We have cleaned the noisy IMDB-WIKI dataset using a constrained clustering method, resulting this new benchmark for in-the-wild age estimation. The annoations also allow this dataset to used for some other tasks, like gender classification and face recognition/verification. For more details please refer to our FPAge paper.

How to use

Clone this repo, install the python requirements and run the script:

pip install -r ./requirements.txt
bash run_all.sh

This will download the original images from the IMDB-WIKI dataset. The file tree would become the following:

data
├── imdb
├── imdb-clean-1024
├── imdb-clean-1024-visualisation
csvs
├── imdb_test_new.csv
├── imdb_train_new.csv
├── imdb_valid_new.csv
├── imdb_test_new_1024.csv
├── imdb_train_new_1024.csv
└── imdb_valid_new_1024.csv

The cropped images are stored in imdb-clean-1024 and the annotations for the splits are in imdb_*_new_1024.csv which you can use to train age/gender estimation models.

Visualisation

Below are samples from imdb-clean-1024:

Community Contributions

• Authors of MiVOLO have kindly provided extra face and body bounding box annotations here.

Disclaimer

We only provide new annotations under MIT licence. The images are from the IMDB-WIKI dataset. We do not own any of these images. Please refer to their website for the licence to use these images.