Am kind of new to deep learning. I have however trained a number of face images and achieved relatively good recognition rate. Now my question is if my client will be taking a few more pictures daily or weekly, how will that be integrated. Or will they have to train again, is there any fast way of adding training data to a pre-trained net?
Although it depends on the library you use, in TensorFlow you can import weights for an existing model and continue training on new (or existing) data. TensorFlow calls this "checkpointing" and I believe it was primarily intended for cases where your model would take too long to train in a single session and you need to pause and resume training later. What you're trying to do is very similar; the only difference is that your data will contain a few more records than before. Reading your new/bigger data set is as simple as just pointing to a different CSV file.
Here is a good example on Stack Overflow of TensorFlow checkpointing: https://stackoverflow.com/questions/33759623/tensorflow-how-to-restore-a-previously-saved-model-python
Step 1: Save the weights
(New data comes in)
Step 2: Make a new dataset (E.g. add new data to the existing dataset or create a completely new)
Step 3: Initialize your model to weights you previously saved
Step 4: Proceed to train on the new dataset (Save the weights after)