Generate Images to Hashtags using CNNs

The notebook dataset_management.ipynb describes the steps to generate the datasets.

It makes use of the fast.ai api to store and verify the images, and it looks for the download links in folder:

• url_files

each class has it's own file, it uses the suffix _valid to indicate the image urls for the validation dataset,

class: autumn

• url_files/autumn.txt
• url_files/autumn_valid.txt

The notebook model_training.ipynb describes the functions and steps to training a model with a custom set of classes.

• Remember that it requires that the Datasets to be already generated in the correct structure.

The notebook predic_classes.ipynb use a generated model and evaluate individual images.

The file app.py has a Flask application that exposes the /image endpoint to process model and generate the probabilites per class.

In the angular-app folder can be found an angular application that can upload images and call the endpoint for the model processing.