Art Appraisal Using Convolutional Neural Networks

The appraisal of artworks has long been an esoteric process reserved for the few educated in art and art history, yet it plays an integral role for collectors, curators, and auction houses. Valuation can be influenced by numerous factors and could be swayed by personal biases and local values. Machine learning algorithms may serve as a tool to adjudicate the value of artwork due to the high subjectivity of the pieces. In order to provide a more standardized method of appraising artwork, we suggest the use of a convolutional neural network (CNN), an algorithm typically used for image classification, to allow an individual without a background in art to better determine the market value of a given piece. We aim for the model to be free of the individual bias that are present in human art appraisers because the dataset is formed from many existing appraisals, averaging over the biases of multiple localities. Our hope is that upcoming artists, art buyers, and art sellers can make more informed decisions with the help of this model, and we may be able to determine cross-cultural features of art pieces that make them inherently valuable.