A5102989985- 3D Shape Modeling and Analysis
- Generative Adversarial Networks and Image Synthesis
- Face recognition and analysis
- Meningioma and schwannoma management
- Medical Image Segmentation Techniques
- Graph Theory and Algorithms
- Advanced Vision and Imaging
- Spinal Fractures and Fixation Techniques
- Human Pose and Action Recognition
- Anatomy and Medical Technology
- Hand Gesture Recognition Systems
- Neurofibromatosis and Schwannoma Cases
- Aortic aneurysm repair treatments
- Advanced Image and Video Retrieval Techniques
- Robotics and Sensor-Based Localization
- Advanced Graph Neural Networks
- Image Processing and 3D Reconstruction
- Advanced Neuroimaging Techniques and Applications
- Neurobiology of Language and Bilingualism
- Spatial Neglect and Hemispheric Dysfunction
- Trigeminal Neuralgia and Treatments
- Surgical Simulation and Training
- Computer Graphics and Visualization Techniques
- Distributed Control Multi-Agent Systems
- Face and Expression Recognition
Laboratoire Angevin de Recherche en Mathématiques
2018-2025
Université d'Angers
2018-2025
Centre Hospitalier Universitaire d'Angers
2016-2024
Inserm
2020-2024
Régulations Naturelle et Artificielle
2024
Université de Bordeaux
2024
University of Bonn
2021-2024
Centre de Recherche en Cancérologie et Immunologie Intégrée Nantes Angers
2024
Nantes Université
2024
Sorbonne Université
2023
We address the highly challenging problem of real-time 3D hand tracking based on a monocular RGB-only sequence. Our method combines convolutional neural network with kinematic model, such that it generalizes well to unseen data, is robust occlusions and varying camera viewpoints, leads anatomically plausible as temporally smooth motions. For training our CNN we propose novel approach for synthetic generation data geometrically consistent image-to-image translation network. To be more...
In this paper, we provide a detailed survey of 3D Morphable Face Models over the 20 years since they were first proposed. The challenges in building and applying these models, namely capture, modeling, image formation, analysis, are still active research topics, review state-of-the-art each areas. We also look ahead, identifying unsolved challenges, proposing directions for future highlighting broad range current applications.
In this work we propose a novel model-based deep convolutional autoencoder that addresses the highly challenging problem of reconstructing 3D human face from single in-the-wild color image. To end, combine encoder network with an expert-designed generative model serves as decoder. The core innovation is differentiable parametric decoder encapsulates image formation analytically based on model. Our takes input code vector exactly defined semantic meaning encodes detailed pose, shape,...
StyleGAN generates photorealistic portrait images of faces with eyes, teeth, hair and context (neck, shoulders, background), but lacks a rig-like control over semantic face parameters that are interpretable in 3D, such as pose, expressions, scene illumination. Three-dimensional morphable models (3DMMs) on the other hand offer parameters, lack photorealism when rendered only model interior, not parts image (hair, mouth background). We present first method to provide pretrained fixed via 3DMM....
The reconstruction of dense 3D models face geometry and appearance from a single image is highly challenging ill-posed. To constrain the problem, many approaches rely on strong priors, such as parametric learned limited scan data. However, prior restrict generalization true diversity in facial geometry, skin reflectance illumination. alleviate this we present first approach that jointly learns 1) regressor for shape, expression, illumination basis 2) concurrently model. Our multi-level model...
Removed.
We propose a method for generating video-realistic animations of real humans under user control. In contrast to conventional human character rendering, we do not require the availability production-quality photo-realistic three-dimensional (3D) model but instead rely on video sequence in conjunction with (medium-quality) controllable 3D template person. With that, our approach significantly reduces production cost compared rendering approaches based models and can also be used realistically...
Monocular image-based 3D reconstruction of faces is a long-standing problem in computer vision. Since image data 2D projection face, the resulting depth ambiguity makes ill-posed. Most existing methods rely on data-driven priors that are built from limited face scans. In contrast, we propose multi-frame video-based self-supervised training deep network (i) learns identity model both shape and appearance while (ii) jointly learning to reconstruct faces. Our learned using only corpora...
We present a novel method for real-time pose and shape reconstruction of two strongly interacting hands. Our approach is the first two-hand tracking solution that combines an extensive list favorable properties, namely it marker-less, uses single consumer-level depth camera, runs in real time, handles inter- intra-hand collisions, automatically adjusts to user's hand shape. In order achieve this, we embed recent parametric model dense correspondence predictor based on deep neural network...
Editing of portrait images is a very popular and important research topic with large variety applications. For ease use, control should be provided via semantically meaningful parameterization that akin to computer animation controls. The vast majority existing techniques do not provide such intuitive fine-grained control, or only enable coarse editing single isolated parameter. Very recently, high-quality controlled has been demonstrated, however on synthetically created StyleGAN images. We...
We present the first deep implicit 3D morphable model (i3DMM) of full heads. Unlike earlier face models it not only captures identity-specific geometry, texture, and expressions frontal face, but also entire head, including hair. collect a new dataset consisting 64 people with different hairstyles to train i3DMM. Our approach has following favorable properties: (i) It is head that includes (ii) In contrast mesh-based can be trained on merely rigidly aligned scans, without requiring difficult...
Tracking and reconstructing the 3D pose geometry of two hands in interaction is a challenging problem that has high relevance for several human-computer applications, including AR/VR, robotics, or sign language recognition. Existing works are either limited to simpler tracking settings ( e.g. , considering only single hand spatially separated hands), rely on less ubiquitous sensors, such as depth cameras. In contrast, this work we present first real-time method motion capture skeletal...
In this work, we propose a novel model-based deep convolutional autoencoder that addresses the highly challenging problem of reconstructing 3D human face from single in-the-wild color image. To end, combine encoder network with an expert-designed generative model serves as decoder. The core innovation is differentiable parametric decoder encapsulates image formation analytically based on model. Our takes input code vector exactly defined semantic meaning encodes detailed pose, shape,...
Synthesizing realistic videos of humans using neural networks has been a popular alternative to the conventional graphics-based rendering pipeline due its high efficiency. Existing works typically formulate this as an image-to-image translation problem in 2D screen space, which leads artifacts such over-smoothing, missing body parts, and temporal instability fine-scale detail, pose-dependent wrinkles clothing. In paper, we propose novel human video synthesis method that approaches these...
The alignment of a set objects by means transformations plays an important role in computer vision. Whilst the case for only two can be solved globally, when multiple are considered usually iterative methods used. In practice perform well if relative between any pair free noise. However, noisy available (e.g. due to missing data or wrong correspondences) may fail. Based on observation that underlying noise-free retrieved from null space matrix directly obtained pairwise alignments, this...
Abstract Introduction Preoperative language mapping using functional magnetic resonance imaging (fMRI) aims to identify eloquent areas in the vicinity of surgically resectable brain lesions. fMRI methodology relies on blood‐oxygen‐level‐dependent (BOLD) analysis areas. Task‐based studies BOLD signal increase during a task areas, which requires patients' cooperation, whereas resting‐state (rsfMRI) allows identification networks without performing any explicit through synchronicity spontaneous...
Tracking and reconstructing the 3D pose geometry of two hands in interaction is a challenging problem that has high relevance for several human-computer applications, including AR/VR, robotics, or sign language recognition. Existing works are either limited to simpler tracking settings (e.g., considering only single hand spatially separated hands), rely on less ubiquitous sensors, such as depth cameras. In contrast, this work we present first real-time method motion capture skeletal surface...
In this work, we introduce the first unsupervised method that simultaneously predicts time-varying neural implicit surfaces and deformations between pairs of point clouds. We propose to model movement using an explicit velocity field directly deform a modified level-set equation. This equation utilizes iso-surface evolution with Eikonal constraints in compact formulation, ensuring integrity signed distance field. By applying smooth, volume-preserving constraint field, our successfully...
Abdominal aortic aneurysm (AAA) is a life-threatening condition involving the permanent dilation of aorta, often detected incidentally through imaging for some other condition. The standard clinical approach to managing AAA follows one-size-fits-all model based on size and growth rate, leading underestimation or overestimation rupture risk in individual patients. widely studied stress-based estimation using computational biomechanics requires wall strength information. However, non-invasive...
Abstract Purpose In orthopaedic surgery, achieving optimal exposure for acetabular and pelvic ring fractures with minimal invasiveness remains a challenge. This study compares bone in key zones using an endoscopic approach versus the AIP (Modified Stoppa) cadaveric specimens. Materials methods We dissected ten adult bodies, obtained from our institution’s body donation program, extraperitoneal dissection on one side other. Bone areas were marked at each step of by drill holes to measure...
OBJECTIVE In recent decades, progress in the medical management of cancer has been significant, resulting considerable extension survival for patients with metastatic disease. This has, turn, led to increased attention optimal surgical bone lesions, including metastases spine. addition, there a shift focus toward improving quality life and reducing hospital stay these patients, many minimally invasive techniques have introduced aim morbidity associated more traditional open approaches. The...