facial recognition in a non photorealistic rendered image and facial recognition across races

[m-hu]

• face_recognition version:1.2.3
• Python version:3.6.8
• Operating System:macOS Mojave

Description

I have tried the followings and it would be good to have:

1. compare similar faces across races
   • expect to have an option to make it recognised
   • unable to get recognised with the current implementation
2. compare human face to a similar cartoon human face
   • expect to have an option to make it recognised
   • unable to get recognised with the current implementation

What I Did

I have tried to compare faces across races, i.e. compare an asian face to a similar caucasian face, an arabic face to a similar caucasian face. It turned out that the similarity is unable to be recognised (face colour and hair colour might be different)

I have also tried to compare a caucasian face to a similar cartoon face. It turned out that the similarity is unable to be recognised.

• The cartoon face seems to have the following properties that lead to the difference
  1. artistical exaggeration, non photorealistic
  2. hair color difference
  3. skin colour difference

[m-hu]

Please note that the issue was created a little bit late in the evening and it is an open ended topic, so the list is not exhaustive which might make the dataset appear to be imbalanced, but in fact it is a sample and the topic is pro racial equity, because the hypothesis is that two faces might look alike regardless of those properties.

In particular, the race combinations are not exhaustive, so I express my sincere apology to the lovely black people, you are not forgotten. Other than that, the property list is not exhaustive either, for example, the texture is not taken into account, like for myself, so I express a sincere apology to myself as well.

[m-hu]

An old story, let me close it.