Awesome

CelebAHQ-Gender

CelebAHQ with annotations on genders (male or female)

Description

Unlike the CelebA dataset, the CelebAHQ dataset consists of 30,000 high-resolution and high-quality images.
However, there are no available annotations like gender, facial expression, or age.

Thus I trained classifiers (ResNet152 and EfficientNet_B0) to distinguish male and female.
The trained classifiers were applied to classify the CelebAHQ dataset.

Nevertheless, the initial results did not seem to distinguish well.
So, I manually inspected the entire dataset.

You can access the dataset here.

The number of images per classes:
Female: 19,057
Male: 10,943

If you find incorrectly classified samples, please report them to me.

Acknowledgement

Original dataset: CelebAHQ.
Resized dataset: CelebAHQ-256x256.