School Projects

Model Inference Proofs
Application of the B method to a model infering algorithm. Semester Lab Project done under the supervision of Akram Idani and Roland Groz

Wavelet Motion Blending
Implementation of the paper Parametric Motion Blending through Wavelet Analysis as a Blender Python plugin with Wilfried Broissart

Bull Gamma 3 in 3D
3D modeling of the Bull Gamma 3 computer for the computer museum Aconit in Grenoble and making it interactive in Godot
Projet GPU
Implementation of a locality sensitive hashing algorithm on images. Parallelized image hashing and computation of distances between characteristic vectors done on the GPU using PyCuda. Done with Alice Sonolet
SIA Course 2
Implementation of a BVH and Xsens sensor data loader and animation player in the render engine from SIA Course 1 + Maya BVH loader C++ plugin with Alice Sonolet, Michael Bleuez and Clément Malleret
SIA Course 1
Implementation of a real-time render engine in GLSL using fragment shading on GPU with Alice Sonolet
Projet GL
Implementation of a compiler in Java to compiles DECA, a subset of Java, into assembly code with Emma Laguier, Ulysse Rigaut, Clément Pigout, Yasser Shimou
Projet C
Implementation of a jpeg decoder in C with Michael Bleuez and Clément Malleret
Personal Projects

VTuber Model V2
refined vtuber rig of for a friend.
Improvements from V1:
- modified dlib facial features detector for more expressivity compared to the stiffer mediapipe face mesh used in V1
- formulation as a simplex problem solved in real-time to find the best blend shape weights for given facial landmarks compared to a set of calibration examples
- communiacte through UDP between mocap script and rig script to spread the load on multiple CPU cores or machines
- 3D illusion using lattices on 2D planes instead of 3D model to look more like state of the art vtubers
- each 2D layer is digitally painted manually in Krita
Improvements from V1:
- modified dlib facial features detector for more expressivity compared to the stiffer mediapipe face mesh used in V1
- formulation as a simplex problem solved in real-time to find the best blend shape weights for given facial landmarks compared to a set of calibration examples
- communiacte through UDP between mocap script and rig script to spread the load on multiple CPU cores or machines
- 3D illusion using lattices on 2D planes instead of 3D model to look more like state of the art vtubers
- each 2D layer is digitally painted manually in Krita

Pix'Art movie
Animated 3D short film made with some friends for fun during associative campaigns at Ensimag

VTuber Model V1
3D model + rig of the upper body for real-time control through a python plugin in Blender