When applying an image classifier as learned throughout lessons 1 and 2, I imagine (in a real businees application) that after training a model, we would need to get new images, classify it, return the prediction to the user and then learn with this image. But as i’m seeing in the lessons, we basically train a whole lot of images, classify the training set and that’s it.
So my questions is:
- How can we use this model to predict upcoming images, in which I assume (again, with no experience) its done one by one, and learn from it? Do we need to fit the entire model again in every prediction? Or my vision of learning from new images is plain wrong and usually that is not how is done?
Perhaps i have the wrong idea on how CNN’s are used in the real market (since i never worked professionally with it), if so, please enlighten me.
This doubt came as I saw a youtube video of a guy programming a self-driven car in GTA V which ran the script for something like a month collecting data and actively learning thus getting better and better basically every minute, if this is a different technique, and it’s not covered in this course, can you guys share some materials in which I can learn better how does this work?
Thanks in advance!