This is the README from my final project from Ironhack's data analytics bootcamp. My project is a neural network that predicts the artist that has done a painting, the dataset was obtained from Kaggle. This dataset contained a list of the 50 most influential artists, my neural network predicts for 11 artists out of the 50 existing in the dataset, because these are the ones that had larger amounts of paintings to train with.
https://www.kaggle.com/ikarus777/best-artworks-of-all-time
ART-ificial intelligence: this project is a mix of art and science, I have trained with almost 3000 artworks my CNN network, using as the pretrained architecture ResNet50. Finally, achieving an 83% of accuracy.
My code needs 2 input arguments, the folder where you have stored the artwork images' and the number of artworks that you want the network to predict
Python, Pandas, Scipy, Scikit-learn, Keras, Tensorflow, Matplotlib and Seaborn. The training of the network was developed in Colab-Pro as the images used where too large to work with in a standard laptop.
The inspiration comes from my personal experience being a person with a non-science background and doing a Data Analytics bootcamp, thus experiencing myself the mix of art and science.
This CNN predicts the author behind an artwork, there are other projects that predict the type of art where each artwork belongs, but my project is focused on predicting the painter and differenciating the style that each painter has.
Requeriments: all libraries described in technology stack and computer with graphical memory to run the CNN
Parameters: p-path_to_images_folder & n-number of predictions
└── project
├── __trash__
├── .gitignore
├── .env
├── requeriments.txt
├── README.md
├── main_script.py
├── p_acquisition
├── p_analysis
├── p_reporting
├── notebooks
│ ├── acquisition.ipynb
│ ├── analysis.ipynb
│ ├── training.ipynb
│ └── demo.ipynb
├── package1
│ ├── acquisition.py
│ ├── plots.py
│ └── demo.py
└── data
├── raw
├── processed
└── results
Next steps: improve actual accuracy, include new painters and their artworks in the network.
Credits: kernel from the dataset in Kaggle (https://www.kaggle.com/supratimhaldar/deepartist-identify-artist-from-art).
Mail: juliafroch@gmail.com Getting help, getting involved, hire me please.