CV
Download the PDF version using the button on the right.
Contact Information
| Name | Mohamed Ebbed |
| Professional Title | PhD Student in Computer Science |
| mebbed@uni-bonn.de | |
| Website | https://mohamed-ebbed.github.io |
Experience
-
2024 - Present Bonn, Germany
Doctoral Researcher
University of Bonn
Working in the Geometry in Machine Learning group on 4D Reconstruction of Deformable Objects from Multi-view RGB videos using Neural SDFs and Gaussian Splatting.
-
2023 - 2024 Munich, Germany
Machine Learning Working Student
Siemens
Working on 3D Object Detection and Segmentation for Autonomous Driving and Robotics from RGB Images and LiDAR.
-
2023 - 2024 Munich, Germany
Teaching Assistant
Technical University of Munich
Teaching Assistant for the Introduction to Deep Learning and Machine Learning courses.
-
2021 - 2022 Munich, Germany
Deep Learning Student Assistant
Fortiss
Building robust object detection models for autonomous driving, and metrics that measure their robustness on test data. Also worked on the visualization of the learned features for Object Detection.
-
2020 - 2020 Cairo, Egypt
Deep Learning Intern
Affectiva
Built a module for real-time Deep Learning model interpretability for facial emotion classification. The module visualizes the parts of the input image that affect the model’s decisions.
-
2019 - 2019 Cairo, Egypt
Deep Learning Research Intern
Valeo
Published 2 research papers at Autonomous Driving workshops at NeurIPS 2019 and ICCV 2019 on Semantic Motion Segmentation for Autonomous Driving with Deep Learning.
Education
-
2024 - Present Bonn, Germany
PhD
University of Bonn
Computer Science
- Specialization: 3D/4D Reconstruction with Neural Fields and Gaussian Splatting
- Supervisor: Dr. Zorah Lähner
-
2021 - 2024 Munich, Germany
M.Sc.
Technical University of Munich
Computer Science
- Thesis: Neural Shading-based Optimization of Neural Implicit Representations
- Specialization: Computer Vision and Machine Learning
- Supervisor: Dr. Matthias Nießner
- Thesis Grade: 4.0/4.0
-
2016 - 2021 Cairo, Egypt
B.Sc.
Cairo University
Computer Engineering
- Thesis: Generating Human Faces from Input Text using GANs
Publications
-
2026 MoAngelo: Motion-Aware Neural Surface Reconstruction for Dynamic Scenes
International Conference on 3D Vision (3DV)
M. Ebbed, Z. Lähner
-
2019 RST-MODNet: Real-Time Spatio-Temporal Moving Object Detection for Autonomous Driving
NeurIPS ML4AD Workshop
M. Ebbed, H. Rashed, A. El Sallab, S. Yogamani
-
2019 FuseMODNet: Real-Time Camera and LiDAR Based Moving Object Detection for Robust Low-Light Autonomous Driving
IEEE/CVF International Conference on Computer Vision (ICCV) Workshops
H. Rashed, M. Ebbed, V. Vaquero, A. El Sallab, G. Sistu, S. Yogamani
Projects
-
3D Reconstruction for Dynamic Scenes from Multi-view Videos (3DV 2026)
Developed a novel framework extending NeuralAngelo for dynamic 3D scene reconstruction from multi-view videos. The method introduces a flexible template-based approach with jointly optimized deformation fields, enabling high-quality geometry tracking and refinement across time. Achieved state-of-the-art reconstruction accuracy on the ActorsHQ dataset compared to existing dynamic reconstruction methods.
- Python, PyTorch
-
Neural Shading-based Optimization of Signed Distance Functions (Master Thesis)
Developed a novel approach for 3D reconstruction leveraging neural fields, facilitating the joint optimization of a neural Signed Distance Function (SDF), material, and lighting using RGB images with known camera poses. Material properties are characterized by a diffuse term and optionally a specular term, while lighting is represented by either an environment map or spatially varying spherical harmonics coefficients. The appearance decomposition into material and lighting enhances reconstruction quality, enabling finer details in the reconstructed surfaces.
- Python, PyTorch
-
ShadedSDF: Volumetric Surface Reconstruction with Appearance Decomposition Constraints
Developed a technique to integrate appearance decomposition constraints (shading cues), leveraging inverse rendering, to refine neural Signed Distance Functions (SDFs). The method enhances state-of-the-art 3D reconstruction methods such as MonoSDF by incorporating the shading cues, improving the quality of the resulting reconstructions.
- Python, PyTorch
-
SparseVoxels: Learning Sparse Implicit Features for Novel View Synthesis
Developed a Deep Learning approach that surpasses NeRF in novel-view synthesis while achieving a 35x speedup. The method optimizes a Sparse Implicit Voxel Grid for the scene using RGB images with known camera poses, then projects the voxel grid onto a 2D plane and feeds it into a rendering Convolutional Neural Network (CNN) to synthesize novel views.
- Python, PyTorch, MinkowskiEngine, COLMAP
-
Inverse Rendering using Neural Implicit Representations
Built a Deep Learning method that reconstructs a relightable mesh from RGB images of an object with known camera poses. Neural fields represent geometry, appearance, material properties, and environment lighting. The method achieves state-of-the-art results, surpassing inverse rendering baselines such as NeRFactor, NeRV, and NeRD.
- Python, PyTorch, COLMAP
-
Real-time Dense Surface Mapping and Tracking
Built a system with CUDA, C++, and OpenCV that performs real-time surface reconstruction and camera tracking for a scene from input depth images.
- C++, OpenCV, CUDA
-
Neural 3D Shape Generation and Completion
Built a Deep Learning method that performs shape completion and generation from an input signed distance function (SDF) grid. The SDF grid is processed by a 3D CNN, and a Graph Convolutional Neural Network then predicts the triangular mesh of the object.
- Python, PyTorch
-
Semantic Motion Segmentation for Autonomous Driving
Built a Deep Learning system that fuses RGB images and optical flow, and uses temporal information, to improve Semantic Motion Segmentation for Autonomous Driving. The project achieved state-of-the-art results and was published at the autonomous driving workshops of NeurIPS 2019 and ICCV 2019.
- Python, PyTorch, TensorFlow
-
Ritratista (B.Sc. Graduation Project)
Built a Deep Learning system that extends StyleGAN2 to be conditioned on a textual description of a face. It also lets the user change the attributes and head pose of the generated face.
- Python, PyTorch