Procedural Stylization

This project was carried out during my 2nd-year internship of the international MOSIG master specializing in Graphics, Vision and Robotics in Grenoble. This internship was conducted at INRIA Montbonnot-Saint-Martin (38330) under the supervision of Joëlle Thollot and Romain Vergne, research professors in the Maverick team..

Principle

This page lists part of the results obtained from the method I worked on. On this page, you will find non-detailed explanations of the method used. For more information, I invite you to read the corresponding thesis.

Thesis

The aim of my project was to find one or more automatic techniques to stylize 3D scenes and thus create images with existing styles, such as watercolor paintings or pointillism. To achieve this stylized result, we render our scene using images—also called 'splat'. To ensure these images correctly follow the movement of the scene and its rendered objects, we attach them with procedurally generated anchor points (in our case using Worley noise).

Input textures for our method:

Position texture

Base Color Texture

Procedural texture (Modified Worley noise)

Outputs of the algorithm

Fractalization

To ensure that the rendering performs well during animation, especially during zooming in or out, we used a fractalization technique of the noise to ensure an almost constant frequency of the procedural noise, regardless of the distance between the camera and the stylized object (see example below).

Without procedural noise fractalization

With procedural noise fractalization

Results

For each of the results, there is an image corresponding to the color map, the image/splat used for rendering the object, and an animation corresponding to the final result.