Under Construction...

Research

Here are some of my research and R&D works in the context of :

  • texture synthesis
  • real-time denoising of realistic rendering
  • GigaSpace - GigaVoxels (sparse hierarchical data structures and cache mechanism on GPU)

Procedural Stochastic Model of Appearance: Texture & Material synthesis

2016-today

 

Role: PhD student in Computer Graphics

Expected End: between december 2020 and march 2021

ScholarshipMESR scholarship (Ministère de l’Enseignement Supérieur et de la Recherche) - 3 years funding/contract

 

Research Lab: ICube laboratory (the Engineering science, computer science and imaging laboratory)

IGG group (Computer Graphics and Geometry)

CNRS (National Center for Scientific Research)

Research director: Jean-Michel DISCHLER

Co-supervisor: Remi ALLEGRE

Employer: Unistra (University of Strasbourg)

 

 

Subject"Statistical and procedural representations for the control and scalability of material synthesis"

Abstract: This thesis proposes new models for the representation and editing of 3D object materials in the context of the modeling of highly detailed virtual scenes, such as natural landscapes or urban environments, for viewing in real time. The 3D engines commonly represent a material by a set of textures associated with the various parameters of a reflectance model. This representation is inadequate in terms of both control and scaling of massive scenes. Moreover, current techniques require very complete example data for the synthesis of textures. In practice, however, this condition is often not feasible. The aim of this thesis is to develop new representations and algorithms of texture synthesis by example passing to the gigapixel scale and offering precise control of materials at multiple levels of scale, including in the incomplete examples.

Targets: CGI, visual effects, video games

Keywords: texture synthesis, texture analysis, statistics, procedural generation, noise, GPU, deep learning, machine learning, material editing, SV-BRDF

 

Technos: C++, OpenGL, glad, GLFW, ImGui, Qt, CMake, UML, SVN, Github, Python, Matlab

 

[Publications]

2020:

[x] "Semi-Procedural Textures Using Point Process Texture Basis Functions"

       P. Guehl, R. Allègre, J.-M. Dischler, B. Benes, E. Galin

- journal: Computer Graphics Forum (EGSR 2020 special issue), 39(4), 2020.

- conference: EGSR 2020 (Eurographics Symposium on Rendering), July 2020, London

   award : Honorable Mention from the Best Papers committee

 

Dedicated website: in progress...

http://pascal.guehl.pagesperso-orange.fr/index.html

  

 

[Publications]

2019:

- Conference Speaker

[x] IG-RV Research Group in : "Procedural Modeling" [Lyon, 4th April 2019]

[x] SIF-doctorants : Séminaire des doctorantes et doctorants en informatique [Paris, 2nd July 2019]

 2017:

- Poster

[x] "Statistical and procedural representations for the control and scalability of texture/material synthesis"

  => "Poster Day", doctoral school MSII (Mathematics, Sciences of the Information and the Engineer), October 2, 2017, Strasbourg, France

 

[Student volunteer]

- PIDS 2020 (Paris Images Digital Summit) [dedicated to VFX industry]

    29 to 31 january 2020, Centre des arts, Enghien-les-Bains, France : student volunteer

- EGSR 2019 (Eurographics Symposium on Rendering) / HPG (High-Performance Graphics)

    8 to 12 july 2019, Strasbourg, France : team leader student volunteer

- Other: PhD student representative at ICube research lab (2016-2020)

   

[Teaching Experience]

University of Strasbourg, France

- Master ISI (Computer Science, Graphics and Imaging) and Bachelor of Science in Mathematics and Informatics

2020-2021: (96 hours)

"advanced programming" (C++) - [M1]

"data structures and algorithms" (C) - [L2]

"IHM" (html, css...) - [L3]

2019-2020: (46 hours)

"3D graphics programming" (OpenGL, WebGL, C++) - [M1] (24h)

"advanced programming" (C++) - [M1] (12h)

- "AI - artificial intelligence" (Python) - [L3] (10h)

2018-2019: (66 hours)

- "3D graphics programming" (OpenGL, WebGL, C++) - [M1] (24h)

- "algorithmics and programming" (Python) - [L1 Informatics] (24h) - [L1 Maths] (16h)

- "english for informatics"  - [L3] (4h)

2017-2018: (68 hours)

- "graphics hardware and softwares" (OpenGL, C++) - [M1] (36h)

- "algorithmics and programming" (OCaml) - [L1] (32h)

 

[Internships (co-)supervision]

Bachelor and Master ISI (Computer Science, Graphics and Imaging) - University of Strasbourg, France

2020-2021:

"Research Internship" - Master [M2]

- ... - [6 months]

"R&D Project" - Master [M2]

- BVH for GPU ray tracing  [OpenGL]  - [150 h]

- GPU Path-Tracing Global Illumination  [OpenGL, compute shader]  - [150 h]

"Research Project" - Master [M1]

- ...

2019-2020:

"R&D Project" - Master [M2]

- Interactive multimedia installation  [C++, OpenGL, procedural terrain synthesis, image processing]  - [150 h]

- raytracing on CPU (physically-based rendering, Monte Carlo) (C++) - [150 h]

- raytracing on GPU (WebGL)  - [150 h]

"Summer Internship" - Bachelor [L1]

- Virtual Galaxy- [7 weeks] (WebGL, javascript)

Arkanoid video game :  - algorithmics, programmation and software design" - [7 weeks] (Python, PyGame)

- Platform/adventure video game - algorithmics, programmation and software design" - [7 weeks] (Python, PyGame)

2018-2019:

"R&D Project" - Master [M2]

- xxx for texture synthesis" - [150 h] (C++)

"Research Project" - Master [M1]

xxx for texture synthesis" - [150 h] (C++)

"Summer Internship" - Bachelor [L1]

- Mixing audio and procedural shape synthesis" - [5 weeks] (Python, Qt, OpenGL)

- Audio-based procedural terrain generation" - [5 weeks] (Unity, C#)

2017-2018:

"Research Project"  - Master [M1]

- "Procedural patterns synthesis on surface in image plane" - [150 h] (C#, Unity, shaders)

- "xxx for texture synthesis" - [150 h] (C++)

 

[Summer/Winter Schools (enrollment)]

2019:

The mathematics of imaging, January 7-11, 2019, CIRM, Marseille (France)

 [convex methods in imaging, practical machine learning, Bayesian methods in imaging, sparsity in imaging]

2018:

GEOMDATA: introduction to geometric data science, Sept. 10-14, 2018 Frejus (France)

[topological analysis, computational anatomy, front and fast-marching evolution methods, variational methods]

 

[Around my PhD work : speaker]

2019:

[x] Demo to students - PhD work (high school) - ICube research Lab : [Strasbourg, 28th March 2019]

[x] “Software corner” - development of scientific computing software at ICube research Lab [Strasbourg, 7th March 2019]

[x] Demo to students - PhD work (bachelor of science) - ICube research Lab  [Strasbourg, 22nd January 2019]

2018:

[x] "What to do after a high school diploma" - ONISEP [Strasbourg, 14th March 2018]

[x] Demo to students - PhD work (bachelor of science) - ICube research Lab  [Strasbourg, 26th January 2018]

 

 


Real-Time Denoising Of Realistic Rendering

2016: 6 months

 

Role : R&D Intern in Computer Graphics

 

Collaboration:

+ AMD (Advanced Micro Devices)

+ Exalt3D (French AMD start-up)

+ LIRIS (CNRS research lab, R3AM team)

Supervisors: Jean-Claude IEHL (LIRIS), Bruno STEFANIZZI (AMD, Exalt3D)

 

Abstract: Rendering realistic images, based on Monte Carlo methods of numerical integration, is able to produce photorealistic images by simulating the physics of light transport. These probabilistic methods, mostly employed in the film industry and virtual prototyping, however, have the disadvantage of a slow convergence rate. The variance of their associated estimators translates as noise in the images. A good visual quality requires a high number of calculations and a very important rendering time. But there is always a residual noise; hence the need to apply filters to clean the images. Recent advances in the field of filtering are focused in the image space, to overcome the complexity of 3D scenes, but the process is still long. This work seeks to analyze how to apply these treatments in real-time, in a manner suited to graphics cards to take advantage of the performance of their massively parallel processors.

 

Keywords: ray-tracing, realistic rendering, Monte Carlo, filtering, noise, optimization, regression, kernel regression, GPU

 

Technos: FireRays, OpenCL, OpenGL, C++, Qt, UML, CUDA, shadertoy, CMake, SVN, NSight, gDEBugger, GPU PerfStudio

 

Publication: "Post-Filtering Of Realistic Rendering"

Internal research report, Master in Computer Graphics, August 2016, Lyon, France


GigaSpace / GigaVoxels

2011-2015: 3 years 7 months

 

Role: R&D engineer in Computer Graphics (lead developer, project technical manager)

 

Research Lab: INRIA (National Institute for computer science and applied mathematics)

Maverick team (Models and Algorithms for Visualization and Rendering, ex-ARTIS, ex-EVASION)

Supervisor: Fabrice NEYRET (research director)

[1] - GigaVoxels - GigaSpace

 

Abstract: GigaVoxels is a ray-guided out-of-core, on demand production, and smart caching library used for efficient 3D real-time visualization of highly large and detailed sparse volumetric scenes (SVO : Sparse Voxel Octree). GigaSpace is a full-customizable out-of-core, on demand production, and smart caching library based on user-defined hierachical arbitrary space partitionning, space & brick visitors, and brick producer using arbitrary data types. GigaVoxels and GigaSpace are indeed the same tool.

 

Keywords: C++, GLSL, Volume Rendering, GPU, Real-Time Rendering, Sparse Representation, NVIDIA CUDA, Voxels, Multi-scale space-partitioning dynamic tree structure, On-Demand data producer, Cache Management, Visitor function marching the data, Lazy Evaluation

 

Technos: C++, UML, CUDA, OpenGL, CMake, Qt, SVN, NSight, gDEBugger

 

 

Speaker / publications:

- "GTC NVidia" (GPU Technology Conference), March 2013, San Jose, California

- "AFIG", Journées de l'Association Française d'Informatique Graphique, November 2012, Calais, France

 

 

Demo:

- "Convergence of cinema and video games: what are the scientific and industrial issues?"  [stand/demo]

=> In’Tech seminar, December 12, 2013, Inria Grenoble, Montbonnot, France

 

Internships (co-)supervision

2014

 ENSIMAG (Grenoble INP) - Master

---- Real-time optical effects in OpenGL using GigaVoxels (glossy reflections, soft shadows, ...) - [2 months, M1]

---- Collision Detection on GPU for real-time applications using GigaVoxels - [6 months, M2]

2013

ENSIMAG (Grenoble INP) - Master 1

---- On-the-fly Voxelization using GigaVoxels - [2 months]

---- RLE compression, data transfert optimization and Depth Peeling multi-pass rendering using GigaVoxels - [2 months]

University of Paris-Est Marne-la-Vallee - Master 2

----- Study hybrid mode : ray-tracing vs volume rendering (cone tracing) - [6 months,]

2012

University of Grenoble (UJF) - Master 2

---- Hypertexture and noise optimization, Graphics and Compute interoperability enhancement using GigaVoxels - [6 months]

 

 

Teaching Experience:

2014 - ENSIMAG (Grenoble INP): "Image" project [M1] - (24h project supervision)

"Art and science: ray-casting for the Hyperlight project"

[2] - RTIGE, Real-Time & Interactive Galaxy for Edutainment

2010-2014

RSA Cosmos company (planetarium designer)

Observatoire de Paris (research center for astronomy) - GEPI department - Stellar and Galactic Physics team

Maverick team (Models and Algorithms for Visualization and Rendering, ex-ARTIS, ex-EVASION) - INRIA

 

Abstract: RTIGE program proposed to address the problem of display, representation and real-time rendering of scenes composed of massively heterogeneous possibly animated elements. This theme was applied to the particular field of the Galaxy.

Funds: 4-year funded research project by ANR (The French National Research Agency) 

Reference: ANR-10-CORD-0006

Supervisors: Benjamin CABUT (RSA Cosmos), Frédéric Arenou (GEPI), Fabrice NEYRET (INRIA)

Websitehttp://wwwhip.obspm.fr/RTIGE/

 

"A virtual tour through our galaxy"

Info: https://www.inria.fr/en/centre/grenoble/news/a-virtual-tour-through-our-galaxy