I am a machine learning engineer at Sensara Technologies, where I work on image/video content analysis using computer vision and deep learning techniques.
At Sensara I've worked on problems such as channel logo recognition, image quality analysis, object detection/recognition, face recognition, deep metric learning, image similarity, on-device machine learning.
I did my bachelors in Computer Science & Engineering at PES University, Bangalore.
- Low compute, completely offline, on-device privacy enabled Channel Logo Recognition part of Smart TVs.
- Worked on everything starting from data collection, preprocessing, cleaning to model selection to train, test and finally deploying.
- Went through the Idea->Experiment->Test loop over 20+ times to achieve production level accuracy. This was due to minute inter-class-similarities in the logos.
- Deployed model on Android platform using Tensorflow Lite. Added support for Over the Air Model updation.
- Worked on automatic configuration of preferred HDMI and TV Operator without any user intervention.
- Once the user starts watching Live TV, the HDMI Source and TV operator is automatically detected using Deep Learning and configured.
- This helps the user to use all features of our Sensy Launcher without manual configuration.
- Worked on integration of Edge AI features such as Channel, Operator Recognition along with features such as Black Screen Detection, local Audio Content Recognition, HDMI State Detection, Landing Page Detection.
- Worked on ensuring lag-free experience for end-user along with these computations.
- Worked on image quality analysis to rank and select best Show Banners, in-Scene, and apparel images. Images were analysed on a range of parameters such as Aesthetic quality, Blurness, Contrast, Interlacing.
- Face Quality Analysis to select the best Faces of people from Scenes and avoided passing bad quality face images for inference to the Recognition Pipeline.
- Improved existing face recognition systems to current state-of-the-art face recognition models.
- This lead to 4x improvement in terms of precision and recall.
- Worked on improving recognition for faces with a higher degree of tilt and occlusions.
- Speeding up of face recognition latency with faster search capabilities.
-> Blog
- Added AI capabilities to the Video Indexer which includes Brand Logo Detection, Object Detection, and Face Recognition for on-Demand videos.
- Includes Face Quality Analysis to showcase the best Face/Banner images used to represent the video.
- Worked on getting the best accuracy, recall with minimal compute and latency.
- Working on automatic matching of Commercials for Products and Programs(House Promos) using unsupervised approaches which includes clustering, distance-based matching.
- Working on Scalable Brand/Product logo recognitions using clustering and deep metric learning.
-> Slides
- Worked on state-of-the-art object detection techniques using Convolutional Neural Networks and Image based Search Retrieval methods.
- Explored at how the recognition speed can be increased by using Locality Sensitive Hashing.
- Innovation Award - Smart India Hackathon 2017 organised by Government of India [News]
- Among Top 10 in Microsoft Mobile Innovation Lab’s Hackathon - 2016
- CNR Rao Merit Scholarship Award for being in Top 20% of CS Dept- PESU - [2014-2018]
Vinay A, D.Naveen Reddy, Abhishek Sharma, Daksha S, N S Bhargav, Kiran M K, KNB Murthy, Natarajan S
ICBDACI, 2017, India
[Paper]
Template credits to Jon Barron.