Deep Learning Face Representation from Predicting 10,000 Classes
read more
Citations
Deep face recognition
A Discriminative Feature Learning Approach for Deep Face Recognition
DeepReID: Deep Filter Pairing Neural Network for Person Re-identification
Deep convolutional neural networks for image classification: A comprehensive review
VGGFace2: A Dataset for Recognising Faces across Pose and Age
References
ImageNet Classification with Deep Convolutional Neural Networks
Gradient-based learning applied to document recognition
Improving neural networks by preventing co-adaptation of feature detectors
DeepFace: Closing the Gap to Human-Level Performance in Face Verification
Labeled Faces in the Wild: A Database forStudying Face Recognition in Unconstrained Environments
Related Papers (5)
Frequently Asked Questions (14)
Q2. What have the authors stated for future works in "Deep learning face representation from predicting 10,000 classes" ?
This could be another interesting direction to be explored in the future.
Q3. How many times do the identity classes increase?
When identity classes increase 32 times from 136 to 4349, the accuracy increases by 10.13% and 8.42% for Joint Bayesian and neural networks, respectively, or 2.03% and 1.68% on average, respectively, whenever the identity classes double.
Q4. How does the neural network achieve the accuracy of the face verification model?
Joint Bayesian only achieves approximately 66% accuracy on these features, while the neural network fails, where it accounts all the face pairs aspositive or negative pairs.
Q5. What is the functionyj in the last hidden layer?
The last hidden layer takes the functionyj = max ( 0, ∑ i x1i · w1i,j + ∑ i x2i · w2i,j + bj ) , (3)where x1, w1, x2, w2 denote neurons and weights in the third and fourth convolutional layers, respectively.
Q6. What is the effect of the bypassing connections between the third and fourth convolutional layers?
Adding the bypassing connections between the third convolutional layer (referred to as the skipping layer) and the last hidden layer reduces the possible information loss in the fourth convolutional layer.
Q7. How many neurons are in the first hidden layer?
It has 38, 400 input neurons with 19, 200 DeepID features from each patch, and 4, 800 neurons in the following two hidden layers, with every 80 neurons in the first hidden layer locally connected to one of the 60 groups of input neurons.
Q8. How many identities are derived from the deep ID?
Highly compact 160-dimensional DeepID is acquired at the end of the cascade that contain rich identity information and directly predict a much larger number (e.g., 10, 000) of identity classes.
Q9. How much accuracy can the authors achieve with Combing 60 patches?
Combing 60 patches increases the accuracy by 4.53% and 5.27% over best single patch for Joint Bayesian and neural networks, respectively.
Q10. What is the simplest formula for the i-th input map?
Max-pooling is formulated asyij,k = max 0≤m,n<s{ xij·s+m, k·s+n } , (2)where each neuron in the i-th output map yi pools over an s× s non-overlapping local region in the i-th input map xi.
Q11. What is the method for detecting facial landmarks?
The authors detect five facial landmarks, including the two eye centers, the nose tip, and the two mouth corners, with the facial point detection method proposed by Sun et al. [30].
Q12. What did they learn from the deep ConvNets?
Sun et al. [31] used multiple deep ConvNets to learn high-level face similarity features and trained classification RBM [22] for face verification.
Q13. How does the transfer learning algorithm compare with the human-level performance of DeepID?
The transfer learning Joint Bayesian based on their DeepID features achieves 97.45% test accuracy on LFW, which is on par with the human-level performance of 97.53%.
Q14. What is the main reason why the classifier outputs are diverse and unreliable?
With so many classes and few samples for each class, the classifier outputs are diverse and unreliable, therefore cannot be used as features.