C2FDAN

We release a dataset to evaluate face identification performance of occluded faces based on MegaFace. Our occlusions vary in form, color, size, and position:

• form: rectangle (aspect ratio 0.5 - 2) and text (list)
• color: random in RGB color space
• size: rectangle and text
  • rectangle: area ratio (area of rectangle / area of image) in {0.05, 0.1, 0.15}
  • text: height ratio (height of text / height of image) in {0.1, 0.3, 0.5}
• position: (mouth, nose, eyes, outside)

Examples of occluded faces with different occlusions:

Download

We provide masks and occluded images for all facescrub gallery images of the Megaface benchmark (DOWNLOAD). The probe images remain untouched. Download the dataset from http://megaface.cs.washington.edu/ or https://github.com/deepinsight/insightface.

MegaFace-facescrub
└───default
│   └───Adam_Brody
│   │   │   Adam_Brody_241.png
│   │   ...
│   ...
│   └───Victoria_Justice
└───square
│   └───eyes
│   │   └───0.05
│   │   │   └───img
│   │   │   │   └───Adam_Brody
│   │   │   │   │   │   Adam_Brody_241.png
│   │   │   │   │   │   ...
│   │   │   │   │   ...
│   │   │   │   └───Victoria_Justice
│   │   │   └───mask
│   │   └───0.1
│   │   └───0.15
│   └───mouth
│   └───nose
│   └───outside
└───text
│   └───eyes
│   ...
│   └───outside

Alignment

To align the faces we make use of 5 landmarks extracted by MTCNN ¹. The alignment script is provided under align_dataset.py, which is based on the code from ².

Cite

If you find PartialLFW useful in your research, please cite the following papers:

@inproceedings{hoermann2021C2FDAN,
    author={Hörmann, Stefan and Zhibing, Xia and Knoche, Martin and Rigoll, Gerhard},
    booktitle={2021 16th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2021)}, 
    title={{A Coarse-to-Fine Dual Attention Network for Blind Face Completion}}, 
    year={2021},
  }

@inproceedings{kemelmacher2016megaface,
  title={The megaface benchmark: 1 million faces for recognition at scale},
  author={Kemelmacher-Shlizerman, Ira and Seitz, Steven M and Miller, Daniel and Brossard, Evan},
  booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
  pages={4873--4882},
  year={2016}
}

References

Contact

Stefan Hörmann ([email protected])

Footnotes

1. K. Zhang, Z. Zhang, Z. Li, and Y. Qiao, “Joint face detection and alignment using multitask cascaded convolutional networks,” IEEE Signal Processing Letters, vol. 23, no. 10, pp. 1499–1503, 2016 ↩

2. https://github.com/davidsandberg/facenet and https://github.com/YYuanAnyVision/mxnet_mtcnn_face_detection ↩