Why is this required?
- While Deep Learning is in many different categories ( like Vision, Text (NLP), Audio, Recommendation system), My interest is always in Vision or anything which involves images, I somehow to find it closer to embed into a hobby than other aspects
- Any Part of Image-based Learning, involves a set of Images that are needed to train the model on what is our intention to recognize parameters, for example, consider the below image set called CIFAR-10.
- https://www.cs.toronto.edu/~kriz/cifar.html — This is the URL, this has a predefined collection of 10 different categories of Images that can be used for training the classifier of an image if the Image is among any of the 10 categories.
- Ordinarily, Let us say you wanted to categorize a Dog in a given picture or you had a scenario where you had months collected time-lapse photos and wanted to categorize Dog and filter out the images which involved Dog in it, you don't have to collect so many images to train, test and build the model, Data is readily available
- Other Popular datasets — https://docs.fast.ai/data.external.html has lot of pre-collected Datasets at our disposal and https://www.kaggle.com/datasets, all the above Data-sets are beginner-friendly Data sets and can be obtained with less of a Hurdle.
Another interesting Method!
- While the above-mentioned Data-sets are the usual ones which we see in the images, the fast.ai course has an interesting approach to bring more data sets and that is via Azure web-API, this seemed so interesting and closer to the real-world that I have seen than any other teaching the course and Fast.ai course implements a ‘bear’ classifier just by pooling images from Azure web-API image collection, details below
- https://github.com/fastai/fastbook/blob/master/02_production.ipynb — Going through this should help setting up an Azure API and fetching images, Fast.ai library makes it even easier to get images, let me show you how easy it is to get images of a bear
