Architectural elements classifier

This project contains the code for the IBM Advanced Data Science Capstone course. You will find thre AI/ML models that predict which is the main architectural element in an image (e.g. a column).

Slides

Dataset

• Architectural Heritage Elements image dataset
• Author: Jose Llamas
• License: Creative Commons Attribution
• 10253 imag
• 10 categories (altar, aspe, bell tower, column, dome…)
• 128x128 color

Models

We have implemented 3 models

• SVM classifier with HOG transformation
• CNN model
• Mobilenet V2 with custom classification output

Usage

• python > 3.8.5
• python package requirements in the requirements.txt file.
• Recommended to have a GPU with cuda and tensorflow configured
• You can run the model_training_routines.ipynb file to train and save the models
• If you want to train them in a more automated way, you can use the classes in model_train and call them from your application

Run the application (use flask only for development purposes):

FLASK_RUN_PORT=8080 FLASK_APP=arch_elements/model_deployment/app.py flask run --host=0.0.0.0

This application implements thre things:

• REST endpoint to receive an image and return the prediction
• Website wher you can upload any image and it returns the predicted element
• Feedback process to help improve the model

License

Architectural elements identifier by 8vicat is licensed under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International