About Me

Muhammad Umair Mukati, Ph.D. - Research Scientist and Machine Learning Engineer

I am Muhammad Umair Mukati, Ph.D., a Research Scientist at GN Store Nord in Denmark, specializing in image processing and machine learning. I am also a Certified SAFe 6 Practitioner.SAFe 6 Certified

I am a highly motivated and results-oriented R&D Engineer with PhD-level expertise and over 10 years of experience in computational imaging, image processing, vision AI, and real-time systems. I have proven ability to develop and implement cutting-edge AI/ML models for image processing and computer vision applications deployed for edge devices.

My expertise spans designing, optimizing, and deploying machine learning pipelines for embedded systems with a strong focus on latency-critical applications and real-time processing. I am passionate about bridging the gap between cutting-edge research and practical applications, with particular expertise in edge deployment, real-time computer vision systems, and cross-functional team leadership in Agile environments.

Technical Skills

Programming & Scientific Computing

  • Python (NumPy, SciPy, Pandas, Scikit-learn)
  • C/C++ (Optimized algorithms, embedded systems)
  • MATLAB (Signal processing, algorithm development)
  • Git, CI/CD pipelines, code documentation
  • Algorithm implementation and optimization

Machine Learning & Neural Networks

  • Deep Learning architectures: CNNs, Object Detection, Segmentation, GANs, etc.
  • Model optimization: Quantization, Pruning, Knowledge Distillation
  • Training approaches: Multi-GPU, hyperparameter tuning, Transfer Learning
  • Deep Learning frameworks: PyTorch, TensorFlow, ONNX Runtime
  • Hardware ML: Qualcomm Neural Processing SDK, NVIDIA Jetson, TensorRT
  • Computer Vision: OpenCV, advanced image processing
  • 3D Computer Vision & Computational Imaging

Signal Processing & Algorithm Development

  • Digital signal processing, real-time algorithm optimization
  • Multi-sensor data analysis and fusion
  • Statistical signal processing, frequency domain analysis
  • Feature extraction techniques
  • Image signal processing, performance optimization
  • Multi-view geometry and computational imaging

Embedded Systems & Hardware Acceleration

  • Resource-constrained and heterogeneous computing
  • Embedded platform optimization
  • Real-time performance optimization
  • Edge AI deployment and acceleration

Data Engineering & Experimentation

  • Data acquisition pipelines, hybrid dataset generation
  • Experimental design, automated testing
  • Performance evaluation metrics and benchmarking
  • Multi-sensor calibration and processing
  • Synthetic and real-world data integration

Software Development & Collaboration

  • Agile methodologies (Scrum, SAFe)
  • Cross-functional team collaboration
  • Technical documentation and knowledge sharing
  • Scientific writing and publication
  • Peer review and collaboration
  • Conference presentations and communication

Professional Experience

Research Scientist: Image Processing and Machine Learning

GN Store Nord | Denmark | February 2022 - Present

Focus: Real-time signal processing, machine learning for edge devices

  • ML Development: Led development of real-time signal processing and ML solutions for resource-constrained edge devices, ensuring high performance with minimal latency
  • Quantization & Optimization: Optimized machine learning architectures for real-time inference through advanced quantization and post-training optimization techniques
  • Benchmarking & Evaluation: Developed and maintained performance evaluation frameworks for ML models, including custom metrics and comparative analysis
  • Prototyping & Validation: Created prototypes for new ML applications, validated results through experimental validation and deployment in production environments
  • Cross-functional Collaboration: Worked with hardware engineers and product managers to align technical solutions with business requirements
  • Technologies: Python, C/C++, MATLAB, PyTorch, TensorFlow, ONNX Runtime, Snapdragon Neural Processing SDK

PhD Researcher: Image Processing and Compression

Technical University of Denmark | Denmark | August 2018 - January 2022

Focus: Multi-view geometry, image compression, light field processing

  • High Fidelity Compression: Developed novel methods for lossless and near-lossless compression of light field images utilizing Epipolar Plane Image (EPI) representations, achieving significant reductions in data size while preserving image quality
  • ML-based Light Field Decoding Developed a low-complexity encoding system using ℓ∞-constrained differential pulse-code modulation (DPCM) combined with a convolutional neural network (CNN)-based decoder. This approach achieved high-fidelity reconstructions with significant improvements over traditional ℓ2-based methods, effectively removing compression artifacts and enhancing visual quality
  • Distributed Compression via View Synthesis Implemented a distributed source coding framework that shifts computational complexity from the encoder to the decoder. By leveraging deep learning-based view synthesis, the system efficiently reconstructed full light fields from a sparse set of views, reducing bitrate by up to 59% compared to HEVC-Intra, thus optimizing storage and transmission efficiency
  • Multi-cylinder Image Representation for Panoramic Rendering Developed a multi-view geometry-based pipeline to generate multi-cylinder image representations using three multiplane images derived from learning-based methods. Enabled efficient dynamic view-point rendering for immersive wide field-of-view visual experiences
  • Technologies: MATLAB, Python, C/C++, PyTorch, TensorFlow

Software Developer

Karachi Institute of Economics & Technology (IGNITE) | Pakistan | August 2017 - February 2018

Focus: Computer vision, face recognition systems, algorithm implementation

  • Implemented CNN-based face recognition system
  • Utilized multiscale techniques to enhance recognition accuracy
  • Applied C/C++ for efficient face localization algorithms

Researcher: Computational Imaging

Medipol University | Turkey | February 2015 - July 2017

Focus: Computational imaging systems, light field processing, optical hardware experiments

  • Multi-view Imaging Systems: Designed computational imaging systems including super-resolution for micro-lens array-based light field cameras
  • Advanced Techniques: Combined captures from conventional photosensors and light field sensors, fused multiple light fields with sub-pixel shifts using mechanical translation stage
  • Optical Experiments: Conducted experiments using Raytrix R10 camera for light field microscopy projects
  • Utilized structure from motion techniques for synthetic enhancement of light field aperture

Founder and Engineer

Jiddat Technologies | Pakistan | January 2013 - January 2015

Focus: Product development, hardware design, embedded systems

  • Product Development: Established company developing affordable pocket-sized microcontroller programmers for AVR and PIC microcontrollers with USB connectivity
  • Full Lifecycle Management: Designed complete product lifecycle including firmware development, PCB design, Gerber file creation, and manufacturing coordination
  • Educational Products: Developed educational kits for control theory courses to facilitate hands-on learning experience
  • Coordinated with manufacturing company in China for PCB production, component soldering, and packaging

Education

Ph.D. in Electrical and Photonics Engineering (2022)

Technical University of Denmark, Denmark

Specialization: Multi-view image enhancement and compression, rule-based and learning-based approaches

Research Focus: Efficient solutions for multi-view image enhancement and compression

Duration: August 2018 - January 2022

M.S. in Electrical Electronic Engineering and Cyber Systems (2017)

Medipol University, Turkey

Focus: Computational imaging methods, multi-view geometry concepts for high-quality image acquisition

Duration: February 2015 - August 2017

B.E. in Industrial Electronics Engineering (2013)

NED University of Engineering & Technology, Pakistan

Focus: Signal processing, control systems, embedded systems

Skills Developed: Industrial controllers (FPGAs, PLCs), embedded tools in industrial settings

Duration: January 2010 - December 2013

Research & Publications

My research interests include computational imaging, vision AI, real-time systems, and machine learning, with a focus on light fields, data compression, and embedded optimization. Below are some of my key publications. For a full list, please visit my Google Scholar profile.


Low-complexity ℓ∞-compression of light field images with a deep-decompression stage

M. Umair Mukati, X. Zhang, X. Wu, Søren Forchhammer

Journal of Visual Communication and Image Representation, 104072, 2024

A novel approach combining low-complexity ℓ∞-based compression with deep learning decompression techniques for efficient light field image coding.

Get In Touch

Location: Bymidten 73F, 2, 3500 Værløse, Denmark

Phone: +45 91 66 50 47

Feel free to reach out for collaborations, discussions, or opportunities.