Software Developer — Advanced Vision based Object Detection with Deep Learning
Robert Bosch GmbH
Oct 2023 – Present | Hildesheim, Germany
Vision-Based Object Detection for ADAS series software
TECHNICAL
- Contributed to the development of next-gen vision based pedestrian detection powered by advanced deep learning, using C++, Python, TensorFlow
- Implemented unit tests using pytest to ensure robustness and reliability of the implemented functionalities
- Optimized detector architecture for hardware compatibility and benchmarked performance through many experiments
- Trained/Finetuned ML models in self-hosted high performance clusters
- Evaluations of the detection and classification quality of the models
- Developed explainable AI tools and visualizations to analyze and understand model decisions and behaviors
- Data pre-labeling with finetuning and inferencing state-of-the-art deep foundation models in Azure, enabling efficient and scalable data annotation workflows
LEADERSHIP
- Facilitated cross-team collaboration to the global teams to define the roadmap to ensure smooth adoption and deployment of next-generation vision based object detection systems
- Led those teams in scaling and integrating advanced object detection features into production pipelines
- Supported global team members in troubleshooting complex issues and optimizing workflows for enhanced efficiency
SKILL
- C/C++
- Python
- TensorFlow
- Git
- pytest
- Linux
- Bash
- Jenkins
- Docker
- Azure
- Computer Vision
- Machine Learning
- Deep Learning
- Advanced Driver-Assistance Systems (ADAS)
Career Break — Parental Leave
Apr 2023 - Sep 2023 | Hildesheim, Germany
Parental leave for 6 months
Software Developer — Vision-Based Object Detection for ADAS series software
Robert Bosch GmbH
Apr 2016 - Apr 2023 | Hildesheim, Germany
Computer Vision feature development for ADAS software targeted at global automotive OEMs
TECHNICAL
- Developed and enhanced ML algorithms for VRU object detection/classification
- Implemented data augmentation features, such as luminance adjustment, geometric distortion, motion blur etc. for a computer vision framwork, using C/C++, OpenCV, OpenMP etc. for core functions, Python for tooling and GoogleTest for test driven development
- Enhanced a Python based evaluation framework with custom metrics (e.g., class/distance-specific ROC, Precision/Recall), tables, visualization, etc. to analyze and improve VRU detection performance
- Conducted extensive experiments and grid-searched hyperparameters to optimize model performance
- Trained, finetuned, maintained scalable vision models for ADAS series software and optimized for hardware deployment
- Implemented toolchain for data classification, selection, clustering, using deep foundation model to reduce the data preparation effort by 80%
- Reduced classifier delivery time by 50% through Jenkins based CI/CD pipelines using Groovy, Python, and Bash
- Developed tools to identify edge cases and false positives, integrating them into retraining workflows
LEADERSHIP
- Guided global teams to adopt new technical workflows, enabling efficient classifier training and deployment in their respective national automotive markets
- Collaborated with stakeholders across multiple cross-functional teams
- Mentored interns, sharing the knowledge of advanced object detection techniques
SKILL
- C/C++
- Python
- TensorFlow
- OpenCV
- OpenMP
- GoogleTest
- Doxygen
- Git
- Linux
- Bash
- Jenkins
- Computer Vision
- Machine Learning
- Deep Learning
- Advanced Driver-Assistance Systems (ADAS)
Computer Vision Engineer — Vision-Based Data Auto-Labelling
CMORE Automotive GmbH
Apr 2015 - Oct 2015 | Lindau, Germany
Vision-based auto-labeling solutions for ADAS applications
TECHNICAL
- Developed an image auto-labeling tool, using C++, OpenCV, and Python, achieving near-human accuracy
- Developed pipelines for the labelling loop, i.e., pre-labelling, manual correction, finetuning ML model
- Optimized label delivery workflows for Tier 1 ADAS suppliers
SKILL
- C++
- Python
- OpenCV
- Doxygen
- Git
- Linux
- Bash
- Jenkins
- Computer Vision
- Machine Learning
- Deep Learning
Master Thesis — Runtime-Efficient SVM for Pedestrian Detection
Continental AG
Nov 2014 - Apr 2015 | Lindau, Germany
RESEARCH
- Developed runtime-efficient SVM models using C++, OpenMP, SIMD intrinsic
- Optimized SVM kernel runtime for resource limited ADAS hardware, reducing computational and memory overhead
- Enhanced detection accuracy for skewed datasets using different sampling techniques
- Authored code documentation in Doxygen and research transcript in LaTeX to ensure maintainability
SKILL
- C/C++
- OpenMP
- SIMD
- MATLAB
- Git
- Doxygen
- LaTeX
- Computer Vision
- Machine Learning
Internship — Vision-Based Pedestrian Detection at Night
Continental AG
Nov 2014 - Apr 2015 | Lindau, Germany
RESEARCH
- Implemented an advanced fast multi-exposure HOG feature descriptor using C/C++ for detecting pedestrians at night scenario
- Implemented data augmentation techniques, such as brightness adjustment, contrast enhancement, noise addition, tone variation etc.
- Evaluated models with metrics like Precision/Recall, Average Precision, FPPI/MissRate, F1-score to ensure performance reliability
SKILL
- C/C++
- SIMD
- OpenMP
- MATLAB
- Git
- LaTeX
- Doxygen
- Computer Vision
- Machine Learning