For the last year of CRISP, we focused our efforts on perceptually-based rendering (with Matin S. Banks), as well as on using machine learning for computer graphics applications (with Alexei Efros).

See more details on ongoing projects (INRIA LDAP password required)


This year we published two papers in the Computer Graphics Forum journal, which were presented at the Eurographics Symposium on Rendering (EGSR). In the first paper we used a model of texture similarity to transfer seasons between photographs. Our algorithm predicts how to change colors and textures in an image to give it the seasonal appearance of another image. In particular, our method captures season-related effects such as leaves on trees, snow and flooding. This work was done in collaboration with Alexei Efros who is an expert in data-driven image manipulation.

The second paper contributes to more traditional, physically-based rendering using bidirectional path tracing. The key idea behind our approach is to exploit combinatorial explosion to cheaply construct a set of light paths as the Cartesian product of the eye and light sub-paths. The novelty of our work is to approximate the contribution of these paths in a probabilistic manner, without constructing each path in the set explicitly. This work results from collaboration with Ravi Ramamoorthi.


The first years of the CRISP collaboration allowed us to gain significant expertise in studying how people perceive and create images, which we used to develop new algorithms for image-based rendering and vector graphics. After a very productive year in 2013, we continued our work on drawing and manipulating materials in vector graphics in 2014.

In particular, we developed a method to assist the conversion of a bitmap image into a vector representation with layers. This representation is well adapted to the capture of realistic material effects like shadows and reflections that artists typically draw with transparent layers. This work was published in the Computer Graphics Forum journal and presented at the Eurographics Symposium on Rendering (EGSR).

While we are planning to continue research on perception and image-based rendering, we started new projects on physically-based rendering and on image editing. These projects represent unique opportunities to collaborate with Ravi Ramammorthi (now at UC San Diego) and Alyosha Efros (who recently joined UC Berkeley), who are experts on these topics.


During the third year of CRISP, our multidisciplinary collaboration resulted in three publications in ACM Transactions on Graphics, the leading journal of our field:

Our goal is now to build on the experimental protocols and methodology that we have acquired through our three years of collaboration. In particular, we plan to conduct experiments to better understand the perception of image-based rendering in more complexe scenarios. We also want to leverage our understanding of the artistic process to analyze existing drawings and facilitate their manipulation.


In 2012 we continued our efforts along the three research directions of CRISP.

  1. Perception.

    We have conducted an extensive perceptual study to evaluate how people perceive materials in non-photorealistic images. Our paper on this topic has been conditionally accepted and is under review after a major revision.

    We have started two other studies on the perception of materials in realistic images and on the perception of facades in image-based rendering. Christian Richardt, who is joining INRIA as a postdoc in October 2012, will focus on these projects for which we were lacking workforce.

  2. Rendering.

    We have extended our algorithm to render objects with rough refraction in real-time. Our method now handles objects with spacially-varying roughness and thin objects lit by local light sources. The extended version of our paper has been published at IEEE Transactions on Visualization and Computer Graphics.

  3. Interaction.

    We are working on novel drawing tools to depict materials in illustrations. Jorge Lopez-Moreno (postdoc INRIA) will visit Maneesh Agrawala (UC Berkeley) in December to work on this project.

    We are exploring the use of phase to control structure in procedural Noise patterns. This project in collaboration with Ravi Ramamoorthi (UC Berkeley) was initiated during the visit of Gaurav Chaurasia and George Drettakis at Berkeley during the summer.


Our first year of collaboration allowed us to make significant progress on the three research directions of CRISP.

  1. Perception.

    We chose to first focus our efforts on the study of material perception in synthetic images. The results of our first study are under submission and we are currently designing a second study on that topic.

  2. Rendering.

    We have developed a new algorithm to render objects with rough refraction in real-time. Our method achieves real-time performances by approximating the complex light transport phenomena involved in rough refractions. Our results are visually similar to images computed with an offline pathtracer, which can be explained by the fact that people have difficulties to evaluate the correctness of complex refractions. The results of this research have been published at the ACM Symposium on Interactive 3D Graphics and Games 2011 where we received the Best Paper Award.

  3. Interaction.

    We have proposed an automatic system to faciliate the design of effective lighting for material depiction. We derived guidelines on lighting design for material depiction from books on photography and studies on material perception. Our system optimizes an environment map to fullfil these guidelines. The resulting lighting emphasizes material-specific visual features. This research has been published at the Eurographics Symposium on Rendering 2011.