Turning Road by P. Cezanne

The Turning Road (Route tournante) exemplifies Cézanne’s extensive exploration of the winding road motif within country side landscapes. The painting is a typical Cezanne, featuring a winding path or road, meandering through a lush landscape, with trees, houses and mountains in the background, rendered with the thick, deliberate brushstrokes and vibrant, yet earthy color palette that are hallmarks of Cézanne’s style. By intentionally leaving parts of the canvas unpainted and integrating these blank spaces into his overall design, Cézanne challenged conventional academic standards. Such defiance cemented his legacy as a forerunner of modernism, earning him posthumous recognition among avant-garde artists of the early 20th century.

Cezanne’s Technique and Color

The painting is an example of Cezanne’s characteristic color palette and brush strokes. It uses a sombre yet harmonious color scheme, with an array of blues, greens, and ochres that hark back to Cézanne’s preferred palette during his time in the Provençal landscape. These colors are layered to build a rich tapestry of composition, while the tactile brushwork—where individual strokes stand out—mirrors Cézanne’s characteristic technique of assembling hues in distinct, patch-like configurations. This treatment of the landscape demonstrates an affinity for geometric abstraction, where forms are distilled into blocks and planes, subtly reducing the illusion of depth and concentrating instead on the structural essence of the scene. The painting’s composition cleverly mediates between flatness and spatial depth, suggesting perspective without fully committing to it, thereby engendering a dynamic interplay between the canvas’s two-dimensional surface and the three-dimensional quality of the depicted world.

Inpainting With Artificial Intelligence

We employed artificial intelligence to simulate inpaintings of the missing portions by taking special care to align with Cézanne’s distinct palette and brushwork. Our aim was to provide inpainting alternatives which closely mirror the artist’s approach to color and texture.

Generating Inpaited Candidates

To inpaint the unfinished segments of the artwork, our AI model self-learned Cezanne’s style and together with his Post-Impressionist school, with the aim to produce a completion faithful to the artist. In addition to the model’s autonomic learning, it was enriched further by textual prompts informed by human expertise. These prompts, carefully curated through a crowdsourced initiative involving artist collectives, injected an element of scholarly insight into the AI’s operation. With a wealth of communal artistic intelligence at its disposal, our AI then synthesized these inputs, employing its algorithms to emulate the stylistic characteristic of Cézanne’s approach.

Our AI model was capable of producing several inpainted simulations of the artwork with high fidelity to Cezanne’s work, as shown below. The images produced by the model are all plausible candidates for an inpainted version of the original work. The next step is to select the best image within the three candidates.

Selecting the Best Inpainting Candidate

As shown above, the model can generate several high-quality candidates suited for the inpainting of the piece. The last step consisted in selecting ona final result that best follows Cezanne’s style in the painting. Our model’s selection process is based on ultimately choosing the restoration outcome that most authentically resonated with the technique and palette that defines Cézanne’s legacy, ensuring the final result was not only a testament to the power of technology but also a reverent nod to the human mastery inherent in Cezanne’s art.

Related Publications

The Artificial Intelligence model is described in our CVPR 2022 publication. The candidate selection model is described in our ICML 2022 publication.