Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Facenet - Paper Review

51 views

Published on

Paper review of the FaceNet paper for the Machine Learning at Microsoft Youtube channel.
https://www.youtube.com/watch?v=w--c0qG9MCc&t=2392s

Published in: Engineering
  • Be the first to comment

Facenet - Paper Review

  1. 1. FaceNet: A Unified Embedding for Face Recognition and Clustering Florian Schroff Dmitry Kalenichenko James Philbin Google [2015] [ 0.01949719, 0.09399229, -0.01618082, -0.00876935, 0.03146157, 0.06853894, 0.00096175, -0.06854118, -0.04771797, -0.05296798, 0.02119147, 0.00511259, 0.1372668
  2. 2. [ 0.01949719, 0.09399229, - 0.01618082, - 0.00876935, 0.03146157, 0.06853894, 0.00096175, - 0.06854118, - 0.04771797, - 0.05296798, 0.02119147, 0.00511259, 0.13726683, - 0.05780432, - 0.00541799, 0.01251621, - 0.08900651, - 0.15897971, - 0.07564467, 0.16574059, …]
  3. 3. EMBEDDING
  4. 4. MOTIVATION PROBLEMS WITH OLD APPROACHES
  5. 5. Is this Tess? This is Anders and Tess? Who diz? VERIFICATION IDENTIFICATION CLUSTERING
  6. 6. TIMTESS TEMPEST YANA TESS UNKNOWN? SIMON CHRIS
  7. 7. We have lots and lots of people…
  8. 8. We need lots and lots of selfies…
  9. 9. FRANCOIS Let’s not hire him, I don’t want to retrain the network TIMTESS TEMPEST YANASIMON CHRIS
  10. 10. This Photo by Unknown Author is licensed under CC BY-NC LET’S TRAIN A SIMILARITY FUNCTION INSTEAD
  11. 11. EMBEDDINGS
  12. 12. APRICOT PEACH BEACH
  13. 13. MOVIE RECOMMENDATION
  14. 14. AdultChild Blockbuster Arthouse * (-1,-0.95) * (0.65, 0.2)
  15. 15. * (-1,-0.95) * (0.65, 0.2) DISTANCE d(P,Q) = f(P)−f(Q) 2 2 𝐿2 𝑁𝑜𝑟𝑚
  16. 16. * (-1,-0.95) * (0.65, 0.2) DISTANCE Embedding on the d-dimensional hypher-sphere f(X) 2 2 = 1 0 = Identical 4 = As different as can be d(P,Q) = f(P)−f(Q) 2 2 𝐿2 𝑁𝑜𝑟𝑚
  17. 17. USING EMBEDDINGS TO SOLVE FACE TASKS
  18. 18. Is this Tess? VERIFICATION Verification = Distance Threshold 𝑑(𝑇𝑒𝑠𝑠, 𝑇𝑒𝑠𝑡) ≤ 𝜏
  19. 19. This is <unknown> and Tess? IDENTIFICATION Identification = search lowest distance or K-nn/SVM classification 0.3 1.2 1.4 2.3 1.7 1.1 1.6 0.8
  20. 20. Who diz? CLUSTERING
  21. 21. DEEPFACE
  22. 22. DeepFace by Facebook 1 0
  23. 23. FACENET TRIPLET LOSS AND TRIPLET SELECTION
  24. 24. TRIPLET LOSS ANCHOR (A) POSITIVE (P) ANCHOR (A) NEGATIVE (N) d(A,P) ≤ d(A,N)d(A,P) - d(A,N) ≤ 0𝑓 𝐴 − 𝑓(𝑃) - 𝑓 𝐴 − 𝑓(𝑁) ≤ 0𝑓 𝐴 − 𝑓(𝑃) - 𝑓 𝐴 − 𝑓(𝑁) + α ≤ 0 0.3 0.32
  25. 25. COST FUNCTION 𝑓 𝑎 − 𝑓(𝑝) - 𝑓 𝑎 − 𝑓(𝑛) + α ≤ 0
  26. 26. FINDING THE RIGHT TRIPLETS We need triplets that violate the equation to ensure fast convergance 𝑓 𝑎 − 𝑓(𝑝) - 𝑓 𝑎 − 𝑓(𝑛) + α ≤ 0 find P where argmax( 𝑓 𝑎 − 𝑓(𝑝) ) find N where argmin( 𝑓 𝑎 − 𝑓(𝑛) )
  27. 27. Triplet selection options A) Generate triplets offline every N steps from a subset of data B) Generate triplets online from a minibatch Paper uses option B, with large minibatches with 1000s of examples
  28. 28. CNN ARCHITECTURE
  29. 29. OpenFace – Open Source FaceNet implementation
  30. 30. GoogLeNet Inception Network
  31. 31. DATASETS PREPARATION AND RESULTS
  32. 32. http://vis-www.cs.umass.edu/lfw/ 13233 images 5749 people 1680 people with two or more images Labeled Faces in the Wild (LFW) 77.5 % Male 83.5 % White (Han & Jain, 2014)
  33. 33. BEYOND THE PAPER
  34. 34. https://www.researchgate.net/publication/317649493_Analysis_and_ Automation_of_Deep_Face_Recognition
  35. 35. DATA PREPRATION with DLIB

×