Learning_Face_Representation_from_Scratch.org

Paper Information

Title

Learning Face Representation from Scratch

Author

Dong Yi, Zhen Lei, Shengcai Liao, Stan Z. Li

Published

Arxiv

Code

None

Paper Content

Why

Sometimes data is more important than algorithm.

What

1. Propose a semi-automatical way to collect face images.
2. Use a 11-layer CNN to learn discrimative representation.

How

IMDB is well-structured.

Data collection

1. extract the feature template of each face by a pretrained face recognition engine
2. use the “main photo” of each celebrity as its seed.
3. use the images contains 1 face to augment each celebrity’s seeding images.
4. for the remain images in “photo gallery”, find the correspondence between faces and celebrities constrained by similarity and name tag.
5. crop face from images and save into independent folder for each celebrity, manually check the dataset and delete the false grouped face images.

Learning Deep Representation

11 layer CNNs gray-scale

Results

CASIA-WebFace has 10575 subjects and 494,414 face images.

Thoughts