Appearance Gender Classifer

NOTE: This model requires cropped face images to work properly.

Purpose

This model classifies a face into female and male appearance.

Architecture

This is a ResNet34 v1 with crop_type='random1', crop_size=[224, 224, 3] and downsample_size=[224, 2 * 224 + 1]. We also used Adam with LR=0.0001 and decrease LR by 0.1 for every 30 epochs.

Intended Use

General purpose gender classifier given a face image.

Limitations

The model might not work very well on small (e.g., less than 50x50) or blurry images.