Pictorial Composition Perception

Artistic composition can be roughly defined as a structural organisation of pictorial elements on a canvas. Even if art history provides basic rules and heuristics, pictorial element's segmentation, interactions and measurements are not well defined. As part of my PhD, these experiments begin to explore the problematic.

Compositional Space

Based on my personal database of small compositions, I have built a deep learning model able to extract and organize the expressivity of my artworks into a high dimensional space: a 'compositional space'. The resulting A.I. object is not only a creative tool to generate novel compositions, but also, a great controlled environnement to run experiments on the pictorial composition perception.

Compositional Similarity

Select the most similar pairs of compositions. About 15min per series.

Recording ending on Sept. 1, 2021

Begin

Painting Orientation

With the hypothesis that a true orientation is the one chosen by the artist, previous experiments discovered that naive and more experimented viewers significantly agreed on the right orientation of abstract paintings (around twice the chance). The following experiments aim at understanding the perceptual mechanisms involved in this judgement by comparing human performances on this task with a machine learning algorithm.

Painting Orientation

Guess the right orientation of abstract paintings. About 1min per series of 11 paintings.

Recording ending on Aug. 11, 2019

Begin
Fragment Orientation

Guess the right orientation of fragments of paintings. About 1min per series of 15 fragments.

Recording ending on Aug. 11, 2019

Begin

Acknowledgements : PSL University (scholarship), ENS (doctoral school), SACRe and LSP (hosting labs).