Mohamed Shawky Sabae

Experience

• 

  Valeo

  Jul 2019 - Sept 2019

  Deep Learning Research Intern

  - Studied the effect of neural style transfer in modeling the real sensor noise, experimented with offline style transfer methods, including single, multiple and arbitrary style transfer. - Applied generative adversarial networks (GANs) like CycleGAN, and neural style transfer on domain translation of LiDAR data from simulated (CARLA) to real (KITTI).- Built a semi-supervised learning problem to train some feature extractor.- Improved the performance of YOLO object detection on LiDAR data by 6%, by augmenting using domain translated data.- Built a framework for testing the quality of the domain translated data.- Built an end-to-end architecture, where YOLO object detector loss in combined with CycleGAN to improve training. - Helped writing and creating the figures of the research paper, which got accepted in NeurIPS 2019 workshop. Show less

• 

  Arete Global

  Jan 2020 - Apr 2020

  Machine Learning Engineering Intern

  - Optimized object detection using RetinaNet architecture for deployment.- Worked on deploying PyTorch models in C++ environments using TorchScript.- Worked on lightweight super resolution neural network architectures.- Worked with Byte Pair Encoding.- Experimented using Kalman Filters for object tracking.

• 

  Google Summer of Code

  May 2020 - Aug 2020

  Student Developer at RoboComp

  - Project title : DNN’s for precise manipulation of household objects.- Implemented and optimized segmentation-driven 6D pose estimation neural network architecture.- Used CoppeliaSim simulator to test pose estimation performance and augment training data.- Developed two new RoboComp components. One for 6D pose estimation from RGBD data using PVN3D network and the other for 6D pose estimation from RGB data using segmentation-driven 6D pose estimation network.- Adapted the previous work of the new Kinova Gen3 arm to PyRep API for fast and robust path planning and grasping framework.- Integrated and tested pose estimation components with the new software architecture (based on DSR (Deep State Representation) and implemented using CRDT and RTPS) for precise manipulation of household objects.- Helped debugging the core C++ API of the new software architecture and improving some features. Show less

• 

  Universidad de Extremadura

  Aug 2020 - Aug 2021

  RoboComp Developer and Mentor

  - Refined pose estimation and grasping pipeline integration with DSR C++ API.- Implemented an object detection DSR agent using YOLOv4 C++ API.- Worked on newer versions of DSR C++ API and improving API documentation.- Tested DSR Python interface.- Expanded RoboComp's tutorial set with DSR and CoppeliaSim tutorials.- Mentored two projects :1) Deep reinforcement learning for precise robot grasping and object manipulation.2) Monocular depth estimation from RGB signals.

• 

  Anovate.ai

  Aug 2021 - Feb 2023

  Computer Vision Engineer

  - Worked on deep learning applications for 3D computer vision, including 3D scene understanding and 3D object reconstruction.- Developed a pipeline for 3D mesh processing and wireframe parsing.- Worked on instance segmentation for satellite imagery.- Deployed high performance deep learning models in production using tools, including NVIDIA Triton Inference Server.- Managed machine learning projects, along with communication with their clients.

• 

  Corporatica

  Aug 2021 - Oct 2021

  Machine Learning Engineer

  - Applied machine learning methods for conversational AI and question answering.- Worked on real-time video segmentation.

• 

  Faculty of Engineering Cairo University

  Dec 2022 - now

  Teaching Assistant

  - Handling course recitations and labs at Computer Engineering department.- Courses:1) Pattern Recognition and Neural Networks.2) Machine Learning.3) Image Processing and Computer Vision.4) Cognitive Robotics.5) Computer Graphics.

• 

  Rembrand

  Mar 2023 - now

  Computer Vision Engineer

  - Developing computer vision and generative AI solutions for virtual product placement.- Working on inverse rendering methods for material and light decomposition from RGB images.- Working on 3D reconstruction pipelines to recover camera parameters and scene geometry from videos.