The Data
Introduction - Data Source
The dataset titled "Best Artworks of All Time" is available on Kaggle and offers a distinct and extensive compilation of artworks created by the top 50 influential artists in history. The dataset is more than just a collection of art. It holds great cultural and historical value, showcasing the development of artistic expression and techniques throughout many centuries. Every artwork in this collection demonstrates the creative and innovative nature of human beings, providing valuable perspectives on various artistic styles, themes, and movements that have influenced the development of art throughout history. The dataset is a valuable resource for individuals interested in art as well as those working in the field of data science. This platform provides a virtual tour of art history, presenting masterpieces from well-known artists like Leonardo da Vinci, Vincent van Gogh, and Pablo Picasso, which will be of interest to those who appreciate art.
The Artworks - All about the Data.
Model Specific Data Usage
Model-1: Generic Classification of the Artist based on Artwork
This model employs classification algorithms to predict the artist behind a given artwork. It utilizes features extracted from the artwork itself, such as style, color palette, brush strokes, etc., to make accurate predictions about the artist.
Data → Artworks of 51 Artists and Fine-tuned to 11 classes of Top 11 artists by number of artworks drawn.
Model-2: Art-to-Description Model using VitGPT Transformer
This model is trained on a Transformer architecture like VitGPT to generate descriptions for artworks. Through fine-tuning on art-related data, it learns to understand the visual content of the art and generate accurate textual descriptions.
Data → Images, Ground-Description generated from the transformer's pipeline. Artist Name, Image ID all of this aggregated to a list of dictionaries
Model-3: Neural Style Transfer/GAN Architecture
This model utilizes techniques such as Neural Style Transfer or Generative Adversarial Networks (GANs) to create new artworks or synthesize existing ones. Neural Style Transfer transfers the style of one artwork onto another, while GANs generate new, original artworks by learning from patterns and styles in the dataset.
Data → Images and Weights from the Neural Network.
Each of these models serves a distinct purpose within the project, encompassing tasks from artist classification and similarity measurement to description generation and artwork synthesis. The use of various machine learning techniques and architectures provides a multi-faceted analysis of the art dataset, enabling comprehensive exploration and extraction of insights.