Closed IrvingShu closed 6 years ago
You can find the results for facial recognition on p. 539, Fig. 4 of the paper.
For me, the use cases were emotion, handwriting, and object recognition, where I was getting about ~35%, ~85%, and ~85% accuracy on cross validation tests with about 6-10 classes. If you are working with more classes, it might be wise to start with Fig. 4 of the paper to get a feel for how the recognition error scales.
Good luck!
Thanks for your answer. I'm working on combining CNN with JointBayers or PLDA, Did you try it with CNN feature?
Not yet, but it is on my to-do list! If you end up going into the PLDA + CNN features direction, I'd be interested in any lessons you pick up along the way!
Expected Behavior
Actual Behavior
Steps to Reproduce the Behavior