Nilesh Kokane
Development
Karnataka, India
Skills
Machine Learning (ML)
About
Nilesh Kokane's skills align with IT R&D Professionals (Information and Communication Technology). Nilesh also has skills associated with Programmers (Information and Communication Technology). Nilesh Kokane has 13 years of work experience, with 8 years of management experience, including a low-level position.
View more
Work Experience
Sr. Software Engineer
Tata Elxsi
May 2015 - May 2016
- Projects 1. Object detection and recognition for video Tasks and Responsibilities: Worked on computer vision algorithms centered on object detection and recognition in a video. Came up with appropriate statistical models for the structured data. Applied conventional computer vision techniques like SIFT, Histogram of Gradients, and edge detection algorithms. Used classification algorithms like SVM for classification of objects. Deployment on Nvidia board for demonstration.
Sr. Software Engineer
Minda Stoneridge
January 2012 - May 2015
- Projects: 1. Advance Instrument cluster: Task and Responsibilities: Designed, developed User Interface for Instrument cluster using Qt Worked on Yocto Eco-System to build the required packages on Freescale board.
Software/Hardware Intern
SPA Instruments
March 2011 - December 2011
- Projects: 1. PLC based Automated machines: Tasks and Responsibilities: Worked on software-in-loo p for testing and evaluating the different modules used in business logic. Optimizing DL models in terms of space and time. Develop strategies for faster and semi-automated annotation using deep Learning
Tech Lead - Machine Learning
Mercedes Benz Research and Development center
June 2016 - Present
- Projects: 1. MPIC- Driver Monitoring System: Tasks and Responsibilities: Responsible for modules such as Hand joints Localization, Facial landmark Localization, Body joints Localization, Detection, Classification, and Head Pose Estimation in images using deep learning techniques like convolutional neural networks (HR-Net and Hourglass Model) and RankPose. Trained custom CNN (Convolutional Neural Network) models, adapting state-of-the-art architectures to achieve the desired accuracy. Utilized GAN's (Generative Adversarial Network) and other Generative models like VAE for in-painting applications: recovering images with sensor markers that are used on subject for ground truth collection. Applied Domain Adaptation techniques to enhance model accuracy by augmenting additional data. Worked on Active learning to fetch out uncertain/hard to localize samples and train the models to make it more robust. Developed models for edge devices to assess their feasibility. Collaborated closely with software engineering teams to optimize models to 8-bit with minimal accuracy degradation. Implemented pose model RankPose and other machine learning techniques for accurate Head pose estimation using 2D IR image. Evaluated performance and accuracy by employing various approaches from Computer Vision, Machine Learning, and Deep Learning, deploying them in-car. Iteratively designed, developed, trained, and evaluated models on real-world scenarios to meet specific use case requirements. Developed Demo Visualization using OpenCV and Qt for evaluating machine-learning models. Collaborated closely with deep learning algorithm experts, incorporating their recommendations into implementation. Interests: Recommender System Robotics Ranking Systems Computer Vision Natural Language Processing Deep Learning Computer Vision Data Structures and Algorithms Projects: 2. Intelligent Car Interior: Tasks and Responsibilities: Worked on hand pose estimation using Res-Net using IR image. Implemented and evaluated gesture detection and recognition of driver/Co-driver using LSTM (temporal) models. Implemented Model-In-loop for ground truth generation, to speed up data-set generation and ease the efforts on human annotators. Volunteer Experience: Worked closely with software team to optimize the model to 8-bit/ fixed point for porting to embedded hardware. Worked closely with camera vendors to get the required image quality Contributor to Hugging Face. includes reducing artifacts in Image, maintaining optimal SNR and Extensively worked on Variational dynamic range, SFR and depth of field etc. Auto Encoder, music VAE using Point cloud generation and evaluation based on Region of Interest magenta platform. Post processing logic for predicted model for signal smoothing.
Education
