How to classify faces with deep learning and transfer learning?
Answers
What do you mean by classify faces? Detect your face or someone elses. Or do you want to do some sort of racial detection (not a racist)? Try to be clear
It depends on what your previously trained model is that you want to use for transfer learning. Regardless of what you're trying to do I'll kind of outline the method.
1]First get a dataset (Face dataset)
2]Get you pre-trained model
3]Perform gradient descent on your model with respect to your new dataset.
If you already have a properly trained CNN for object or face detection. It won't take long to acheive a desired accuracy but if you have a network trained to do something completly different like classify numbers it will take a long time and be quite ineffecient.
The whole point of transfer learning is to use a pre-trained model and train it to get to the desired accuracy more quickly than you can by training with random weights from scratch. So, while choosing you're pre-trained model be careful. You would not want to take a model that is trained for a completetly different task.