![]() Moreover, we carry out a benchmark experiment by testing eight mainstream backbones on the proposed FDEA. Finally, our FDEA contains 157,801 samples and is divided into three classes: Caucasian (54,438), Asian (61,522), and African (41,841). To this end, we employ nine annotators to label these samples from CelebA, while cleaning the remaining samples manually. The samples extracted from CelebA are not labeled with ethnicity attribute. ![]() For this purpose, we first collect an initial face dataset from CelebA and LFWA, MORPH, UTKFace, FairFace, and the web. This paper proposes a new Face Dataset with Ethnicity Attribute (FDEA), intended for ethnicity recognition benchmark. ![]() This is partly because there is no large enough dataset and labeled accurately with ethnicity attribute. The ethnicity attribute is precious due to its invariance over time, but has not been developed well. However, existing face attributes (such as expression, age, and skin color) are subject to change. Face attributes play an important role in face-related applications.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |