
LVSegmenter: Left Ventricle Segmentation
2025 · Télécom Paris · Medical Imaging · Image Processing
Custom model for left ventricle segmentation in cardiac MRI. Based on a Kaggle-style challenge (IMA205: Apprentissage pour l'image et la reconnaissance d'objets).
Télécom ParisMedical ImagingImage Processing

Figurative vs Abstract Painting Classification
2025 · Télécom Paris · Computer Vision · Deep Learning
Classifying paintings into figurative vs abstract using ResNet50 variants (single-branch, two-branch, dense fusion). Best model achieved ~90% macro F1-score.
Télécom ParisComputer VisionDeep Learning

Prosody-Driven Facial Expression Prediction
2025 · Sorbonne Université · Multimodal Learning · Deep Learning
Transformer model that predicts upper-face Action Units (AUs) directly from speech prosody (F0). Built a full multimodal pipeline with OpenSmile + OpenFace alignment and a multi-decoder architecture for AU generation.
Sorbonne UniversitéMultimodal LearningDeep Learning