Research Reports and presentations

In this page you will find some reports on various topics but also presentations of academic papers or personal work. In particular, presentations are not always explicit without the speaker's explanations, but I think they are still interesting to share as they give a good overview of the covered topic and show some of the work I have done on it.

Medical AI2026

Pain Signal Analysis from Gait Data

A try to characterize the presence of pain signals in the gait signal of injured persons.

Abstract:

In this work, we tried answering a question from a doctor: is there any pain signal inside of the gait signal of injured persons. This presentations summarizes our approach, and the results we obtained.

Time SeriesSignal ProcessingVAERepresentation Learning

3D Computer Vision2026

3D Gaussian Splatting

A comprehensive analysis of 3D Gaussian Splatting, a novel technique for representing and rendering 3D scenes using collections of 3D Gaussians.

Abstract:

This report is an summary of the 3D Gaussian Splatting method, detailing its principles, implementation, and contributions. Concepts are illustrated through 2D studies and limitations of the method are highlighted.

Gaussian SplattingRenderingComputer VisionPython

7 pages2.1 MB

Download

Point clouds and 3D2026

Point Transformers V3

An implementation from scratch of Point Transformers V3 adapted for the Blackwell architecture, a state-of-the-art model for point cloud processing and 3D understanding tasks.

Abstract:

This work presents a detailed analysis of the Point Transformers V3 architecture, including its design, reimplementation and adaptation designed for the new NVIDIA Blackwell architecture, and performance on various point cloud processing tasks. The report also discusses the use of Triton for efficient GPU computation and highlights the model's contributions to 3D understanding.

TransformersTritonPoint CloudsPython

8 pages2.5 MB

Download

3D Computer Vision2026

Zero123++

A presentation of Zero123++, an advanced neural network architecture for generating 3D models from single 2D images, showcasing its capabilities and applications in 3D reconstruction.

Abstract:

This presentation introduces Zero123++, a cutting-edge Diffusion Model designed for generating 3D models from single 2D images.

3D ReconstructionComputer VisionDiffusion Models

PresentationInteractive

Graphs and Retrieval2026

Reproducibility Study: Semantic Image retrieval via Scene Graphs

A reproducibility study of the paper 'Semantic Image Retrieval via Scene Graphs' which proposes a method for retrieving images based on their semantic content using scene graphs.

Abstract:

This report presents a reproducibility study of the paper 'Semantic Image Retrieval via Scene Graphs'. The original paper proposes a method for retrieving images based on their semantic content using scene graphs. In this study, we re-implement the proposed method, evaluate its performance on the same datasets, and analyze the results to assess the reproducibility of the original findings. Challenges encountered during implementation and potential discrepancies in results are also discussed.

GraphsCRFImage RetrievalPython

11 pages0.25 MB

Download

Computer Vision2025

Semantic StyleGAN

Research report on giving semantic meaning to StyleGAN2 latent space dimensions. Explores techniques for controlled image generation and latent space manipulation.

Abstract:

This report presents a methodology for adding semantic interpretability to the latent dimensions of StyleGAN2, enabling more controlled and meaningful image generation.

GANsStyleGAN2Computer VisionPython

20 pages0.72 MB

Download

Medical AI2025

Pathology Prediction from MRI

Comprehensive study on machine learning approaches for medical pathology prediction using MRI data. Achieved 2nd place in a competitive academic challenge.

Abstract:

Implementation of advanced machine learning techniques for automated pathology detection in MRI scans, with focus on accuracy and clinical applicability.

Machine LearningMedical ImagingMRIClassification

11 pages1.5 MB

Download

Medical Imaging2024

3D Cardiac MRI Segmentation

Technical report on classical computer vision approaches for segmenting 3D cardiac MRI images without deep learning techniques.

Abstract:

Exploration of traditional computer vision methods for accurate 3D cardiac MRI segmentation, demonstrating effective results without neural networks.

Computer VisionMedical ImagingSegmentationClassical Methods

9 pages2.3 MB

Download