Test Detect Face Rectangle

[HevaWu]

Overview

Try to implement face detection by using

• Vision
• MLKit

And compare the result of them.

Description

We could use either Vision or MLKit to implement this part. After comparing, I will attach the result here.
The mainly comparison are:

• face detection number
• execution time

Result

• Execution Time:
  • For most cases, MLKit is faster than Vision Framework. As far as I test for now, the execution for Vision or MLKit is <= 0.6s. **
• Face Detection Number:
  • It decide case by case. Vision could quickly fetch the frontal face. But MLKit could also detect more faces when people are wearing mask/hat.
• Image Size:
  • Large Image: Both Vision and MLKit could processing the large image(ex: I pick one of image is 3840*2560, and the processing time is: Vision(0.534s), MLKit(0.409s)).
  • Small Image: The smallest image I pick here is 195*259, which could be processed for both Vision(0.05s) & MLKit(0.03s).
• Take Photo:
  • Both Vision & MLKit could handle the token picture.
• Accuracy:
  • I only use the default settings for Vision & MLKit for now. If the person picture is frontal face, the result is almost same, sometimes Vision could detect more, sometimes MLKit could detect more. But if the person picture wearing some decoration (ex: mask, hat, glasses), MLKit have higher accuracy.

Notice

This testing result is only based on the face frame part, which is not related to other facial landmarks part(the result might different if we try to detect different facial parts).

Reference

Vision: https://developer.apple.com/documentation/vision
MLKit: https://firebase.google.com/docs/ml-kit/detect-faces

Test Result