2
1
Fork
You've already forked ddc
0
5 Home
ojrandom edited this page 2024年11月21日 16:58:42 +00:00

Comparison to other Programs

Please feel free to correct this comparison and contribute the result of own tests!

The test was done with Deepface. Digikam and the Darktable addon might use different implementations. At least the results should point into a direction.

What was tested?

81 pictures

  • 23 x Brigitte Bardot (3 x are named)
  • 51 x Claudia Cardinale (3 x are named)
  • 7 x Marylin Monroe (2 x are named)

Used metric: Cosine with the optimized default thesholds provided by Deepface

  • dlib: 0.07
  • OpenFace: 0.1
  • Facenet512: 0.3
Program Detector Model Pictures Detected Named Recognized False Positives
face_recognition (Darktable) dlib dlib 81 110 8 17 (of 73) 1
Digikam yolov8 *1 OpenFace 81 176 (?) 8 0 (of 73) *2 0
Darktable's Face Companion yolov8 Facenet512 81 176 8 51 (of 73) 3

*1... Digikam uses the outdated yolov3 instead of yolov8.

*2... OpenFace resulted in many false positives if the threshold was increased to recognize more faces.

Summary Darktable's Face Companion Compared

Pros

  • Detection and recognition more up-to-date
  • Variety of detectors and models available
  • Change the method of detection
  • Change the method of recognition
  • Change the distance metric
  • Upgrade of detectors and models if new ones are published (and integrated by Deepface)
  • training data can be filtered and sorted by
    • confidence (this is a face)
    • width of faces relativ to width of picture
    • sharpness
    • how directly a person looks into the camera
    • most similar faces if there are (too) many faces

Cons

  • Not integrated in Darktable

Digikam

Digikam comes with a face detection and recognition, see

Pros:

  • Integrated in Digkam
  • Detection and recognition applied to current collection
  • Training data can be filtered by
    • noise
    • sharpness
    • compression
    • overall quality (AI feature ?)

Cons:

  • Both detection and recognition seem to be quite outdated.

Dartktable's Addon face_recognition

Darktable has the addon face_recognition.

It seems to use

Pros:

  • Integrated in Darktable

Cons:

  • Both detection and recognition seem to be quite outdated.
  • Results can not be confirmed.