About Me
I am a graduate researcher in Computer Science at Keio University (HVRL) and Ecole Centrale de Nantes, with a theoretical ML foundation from PSL University (M2 IASD). My research interests lie in computer vision, particularly learning task-relevant visual representations that generalize across diverse data distributions under limited supervision. I am especially interested in self- and semi-supervised learning and domain generalization, with a growing focus on biomedical imaging. My current research includes particle picking in cryo-electron microscopy, while my previous work has explored robust visual perception, monocular 3D reconstruction, and low-light image enhancement.
CVInterests
- Computer Vision for Biomedical Imaging
- Representation Learning
- Domain Generalization and Robustness
- Self- and Semi-Supervised Learning
- Event Spotting
- 3D Scene Understanding
Education
M.Sc. in Computer Science
Master 2 IASD (M.Sc. AI)
Diplôme d'Ingénieur (BSc & M.Eng.)
French Preparatory Classes (PCSI-PC*), Undergraduate
Publications
Research Experience
Graduate Research Assistant
Cryo-EM Particle Picking (Master's Thesis): Investigating deep unsupervised and semi-supervised learning methods for particle picking in cryo-electron microscopy. The project addresses the challenges of extreme image noise, limited expert annotations, and generalization across diverse datasets.
Volley3D (First-Author Manuscript): Developed a framework for 3D volleyball-scene reconstruction from monocular broadcast footage without 3D supervision. Combined jump-aware player kinematics with 2D-to-3D ball trajectory reconstruction using TCN-based segmentation, collision matching, and a physics-based solver.
PoseConvSpot (Accepted for Oral Presentation at ACM MMSports 2026): Designed a pose-only spatiotemporal regression architecture using Strip Pooling, GRU, and Soft Gamma Loss for precise jump-event spotting. Improved mAP@3 by 14.1% under cross-arena evaluation compared with state-of-the-art RGB-based baselines.
Computer Vision Research Intern — Master's Thesis
Investigated unsupervised image-to-image translation to synthesize nighttime training data from daytime imagery, addressing data scarcity in low-light vision. Enhanced CycleGAN with DINO-based structural priors to preserve scene structure during domain translation, alongside an RGB-to-RAW unprocessing pipeline.
Designed a dual-task evaluation protocol assessing synthetic domain transfer for both low-level image restoration (denoising metrics) and high-level perception (object detection mAP).
Computer Vision Research Intern
Investigated monocular 3D reconstruction of volleyball scenes from uncalibrated broadcast footage, combining player tracking, 3D human pose estimation, and kinematic modeling.
Developed an initial physics-informed approach to handling jump-related ambiguities, laying the groundwork for subsequent research on pose-based jump-event spotting.
Machine Learning Research Project
Contributed to a clinical machine learning project predicting lung transplant rejection from patient data, assisting a doctoral researcher.