So i was playing around with person identification dataset and it doesn’t has classes, so functions like get_data() can’t be used directly. My dataset looks like in the picture above. How do i process it so that i can use it with fastai afterwards ? This can further be extended to other datasets as well.
What do you want to learn? When you know that, you can define your classes by putting them in different directories for example.
I want to identify persons, I am not sure on which basis I should do that ? My concern is how do i process on data using fastai lib in which classes are not known just like this one. There are many datasets that don’t specify classes like object detection or identification.
Please don’t post the same thing to multiple categories.
Apologies ! I wasn’t able to delete that post. Could you please shed some light on how do I do this @jeremy ?
I’ve done it
I checked the whole dl1 course, in dogs-cats, CIFAR10, planet-competition (kaggle) classes were well defined. I am not sure which exercise should I be looking for.
Fast.AI provides ways to build State of the Art models for Supervised Learning Tasks. In Part 2, Jeremy might talk more about semi-supervised learning or other tasks.
So, I would suggest that you use the library for tasks similar to what was covered in Part 1 Lectures -Images, Text and Structured/Tabular data that have labels/classes (Supervised Learning) as of now.
For your problem I would suggest that you spend an afternoon or a few days to manually annotate some of your data with labels. You could then use the fast.ai library and use the existing dl1 notebooks as reference.