About this project:
This project explores the use of machine learning techniques in identifying the genre of fine art paintings.
It is divided into six parts:
- Part 1 Data Collection and Preparation:
- Fine art images were downloaded and separated into different folders according to their genre label
- The images were divided into training, validation, and test sets in preparation for modeling later on
- Part 2 Transfer Learning:
- Pre-trained Convolutional Neural Network (CNN) models such as AlexNet, ResNet-50, and VGG-19 were used
- Applied transfer learning on the pre-trained models by modifying their final layers and training them to classify fine art paintings based on their genre
- Part 3 Custom Model:
- Implemented a CNN model from scratch to classify fine art paintings based on their genre
- Part 4 Visualization of Statistics:
- Compared the performance of pre-trained models and custom model and visualized their metrics
- Part 5 Image Segmentation:
- Experimented with segmenting objects within a fine art painting using a pre-trained image segmentation model such as DeepLab V3
- Applied transfer learning on pre-trained image segmentation model to improve performance
- Part 6 Web App:
- Deployed the best-performing classifier model as a web application
The classifier has been deployed as a web application.
The GIF below demonstrates how to use it:
The web app is hosted on Streamlit Cloud. 
- Set up the environment using Conda:
conda env create -n fine-art-classifier -f environment.yml-
Execute the Jupyter notebooks in the folder
notebooks. -
Create a separate environment for Streamlit app:
conda env create -n fine-art-classifier-app -f app/environment.yml- Launch Streamlit:
streamlit run app/streamlit_app.py