CV

Research Experience

Layer42 · Jan 2025 – Present

Investigating visual place recognition with DINOv2, CLIP, and SigLIP, comparing LoRA, IA3, and adapters alongside SALAD, BoQ, and query-based feature aggregation.
Developed EO-VLM for reproducible evaluation of vision-language models and Terra-Q for onboard image analysis and downlink prioritization, with LoRA fine-tuning of compact VLMs.
Evaluated 2D feature lifting for text-driven 3D querying and benchmarked Gaussian Splatting SLAM methods for tracking, reconstruction quality, runtime, and robustness.

TTI/ATRC · Sep 2021 – Oct 2024 · Abu Dhabi, UAE

Led the development of a UAV visual positioning system combining image retrieval, local feature matching, geometric verification, and pose estimation for aerial-to-satellite localization.
Built georeferenced training data with synthetic flyovers and appearance augmentation; evaluated retrieval with Recall@K and mAP, alongside correspondence stability and pose accuracy.
Developed automated annotation, detection, and tracking pipelines for aerial imagery; optimized inference on NVIDIA Jetson using TensorRT, CUDA, and mixed precision.

Institute of Computer Graphics and Vision, TU Graz · Jan 2020 – Jul 2020 · Graz, Austria

Designed and calibrated a synchronized multi-camera RGB-D setup for overhead detection and tracking, with consistent 2D/3D annotations generated through geometric reprojection.
Developed a tracking-by-detection pipeline combining instance segmentation, appearance descriptors, and spatiotemporal constraints; evaluated detection and identity tracking on MOTS Challenge datasets.

Institute of Computer Graphics and Vision, TU Graz · Apr 2019 – Dec 2019 · Graz, Austria

Co-developed DeepPCF-MVS with Sony Europe to improve planar completeness in COLMAP multi-view stereo reconstructions.
Combined learned plane segmentation, geometric optimization, and robust plane fitting to complete missing depth and surface normals.
Evaluated reconstruction accuracy, completeness, and F1 on ETH3D and Tanks & Temples. DeepPCF-MVS achieved 88.10% F1 at a 2 cm tolerance on the ETH3D high-resolution multi-view test benchmark.

Siemens PLM LMS · Apr 2018 – Sep 2018 · Leuven, Belgium

Developed an RRT-based kinodynamic planner for dynamically feasible autonomous-driving trajectories.
Compared nonlinear model predictive control (NMPC) and combined feedforward–NMPC controllers using ACADO; evaluated trajectory feasibility and tracking stability in Siemens Amesim.

EO-VLM: Reproducible VLM benchmarking on satellite and aerial imagery, covering classification, visual question answering, counting, spatial reasoning, and temporal analysis. Project page
ImMatch: A common interface for local feature extraction, image matching, and geometric estimation.
Mappero: Modular photogrammetry tools integrating COLMAP, GLOMAP, and learned feature matching for reconstruction and mapping.
Retrieval and VisLoc: Research tools for visual place recognition and image-based localization in 3D maps.
DeepPCF-MVS: Plane-aware completion for multi-view stereo, with public reconstruction benchmark results.
NN-MPC: Motion planning and predictive control implementations for autonomous driving.

M.Sc. in Control, Computer Vision, and Signal Processing
CentraleSupélec and Paris-Sud University · Oct 2018 – Dec 2019 · France
Thesis focus: geometry-aware 3D scene completion from multi-view imagery.
M.Sc. in Smart Aerospace and Autonomous Systems
Paris-Saclay University · Sep 2017 – Jul 2018 · France
Thesis focus: motion planning and control for autonomous driving.
M.Sc. in Control and Systems Engineering
INELEC, Boumerdes University · Sep 2013 – Jul 2015 · Algeria
B.Sc. in Electrical and Electronic Engineering
INELEC, Boumerdes University · Sep 2010 – Jul 2013 · Algeria

Geometry and vision: Visual localization, image matching, SfM, MVS, SLAM, Gaussian Splatting, depth estimation, detection, and tracking; OpenCV and COLMAP.
Machine learning: PyTorch, Hugging Face, PEFT, vLLM, vision foundation models, VLMs, and representation learning.
Robotics and control: ROS, RViz, MoveIt, motion planning, model predictive control, and ACADO Toolkit.
Research engineering: Python, C++, MATLAB, Bash, Git, Docker, and Linux; CUDA, TensorRT, ONNX, and NVIDIA Jetson.