This is an AI classifier which uses CNN architecture to classify differents types of fruits. This README only serve as an overview of this project. For a more detailed documentation, checkout the documentation directory, with versions in english and spanish.
In an era where transformers have taken over the AI industry, previous AI architectures can be seen as stone age technology, however, I must argue that that doesn't make other architectures useless.
Think it in this way: If there is a single boulder in the middle of a road, and need to brake it first to transport and process it, whould you bring an entire tunnel excavator? Transformers aren't cheap, so it comes to a "bring the right tool to solve the specific problem, not a silver bullet one".
CNNs are pretty good catching independent and short distance data relations. Which is already what we need to classify a single fruit in a picture. And CNNs are cheaper, images with lower resoluation and a few convoluational and normal/dense layers can be enough to get good results in minutes of training instaed of hours.
The project is separated in three main modules or components:
- model_training: Tells in detail the training process, including an analisis of problems found during training and posible solutions.
- flask_server: Contains a server.py and a client.py scripts to mount a simple client-server application.
- streamlit_app: Contains a streamlit_app.py script to mount the app in an streamlit environment or platform.
As explained in the Notice file, under the given conditions, you can download and use the dataset as you want. But I must warn you that not all the images are different fruit individuals. A few numbers of fruits were photographed, only changing the angle or scenary to increase the dataset. Don't ask me how to differentiate or set appart the individuals, I don't know neither since it wasn't my intention.

