User details
Kamlesh Kumar
@kamlesh_kumar_djgrevzlg
Student
South San Francisco, CA, United States
2Followers11Following0Mutual0Post
Member since: Jan 21, 2025Research Engineer pursuing masters at SJSU looking for AI/ML Engineer roles, focused on Image and Video Under- standing research problems. Brings practical experience as Research Engineer Intern at Adobe contributing to conferences such as CVPR'25, in addition to two patents and research papers. Expertise in implementing and optimizing advanced deep learning solutions using PyTorch, TensorFlow, OpenCV, NumPy, and ISAAC-ROS.
• Resolved problems related to Chroma Keying and Green Screen matting to capture training and eval. dataset. . Collected and curated video matting dataset (360K+ frames) with polarization cameras to estimate opacity.
• Retrained SOTA matting methods, based on- trimaps, masking, background matting, and GenerativeAI with trans- formers and attention models, with upto 50% MSE reduction on public matting evaluation datasets - CVPR'25. . Leveraged AWS S3 buckets, RunAI, Cloud Computing, PyTorch, TensorFlow, Research, Publication, LaTeX.
• Developed calibration and fusion algorithms for 2 sensors, LiDAR and Camera, for 3D perception with ROS2.
• Designed and implemented object detection pipelines to 100% automated lines for robotic process automation.
• Documented research findings and published 2 research papers to strengthen scientific community through open-source. . Frameworks: ISAAC-ROS, C++, PyTorch, TensorRT, ONNX, OpenCV, Docker, Matlab, Linux, Git.
• Designed and developed of over 10 impactful projects in Industrial Automation Solutions, harnessing Computer Science, GenAI, Computer Vision, Programming Language - Python, and Product Management.
• These initiatives optimized security measures up-to 40%, streamlined traffic management, and elevated overall mass production infrastructure.
• Conducted in-depth modelling, analytics, and validation of real-time results across all product deployments, resulting in an impressive 90% reduction in false positive rates.
• Revamped all existing Object Detection/Classification models by implementing accurate and up-to 30% faster algorithms, utilizing cloud computing and GPU training for enhanced performance.
• Demonstrated proficiency in 10+ demanding frameworks, including PyTorch, TensorFlow, Keras, OpenCV, NumPy, SQL, Hadoop, Google Cloud Platform, while showcasing leadership, communication, presentation, and teamwork capabilities.