[Img : Aditi Shanmugam.]
I graduated with a Bachelor's degree in Electronics and Telecommunications Engineering from Visveswaraya Technological University in July '22. During my undergraduate journey, I had the privilege of gaining valuable professional experiences that significantly influenced my career path in the field of Robotics. I began my journey as a Data Science Fellow at Fellowship.ai, where I immersed myself in the intricacies of Machine Learning and AI, acquiring valuable skills and insights. I then had the opportunity to work as a research intern at the Visual Computing Lab within the Department of Computational Data Sciences at the prestigious Indian Institute of Science, Bangalore. This role not only broadened my horizons but also deepened my understanding of advanced research in the field of Deep Learning. A brief yet impactful tenure as a Computer Vision intern at Inferigence Quotient, a trailblazing defense tech startup where my contributions were well-received, led to a full-time position as a Computer Vision Engineer, a role I currently hold. These diverse experiences have equipped me with a solid foundation in Electronics, Telecommunications, and Data Science that renders me exceptionally proficient as a robotics engineer.
Work Experience
Computer Vision Enginner, Ex-Intern
April '22 - Present
I was previously an Intern (April - July '22) and currently part of the Computer Vision Team at Inferigence Quotient as a Computer Vision Engineer, working on development of in-house data annotation tools, overlooking curation of custom datasets, training neural networks to perform Image and video analytics for real time surveilance with autonomous UAVs.
Primary Tasks and Responsibilities:
1. System for Tracking And Recognition of Targets - iSTART- Developed a robust object recognition and tracking pipeline to be deployed within Unmanned Aerial Vehicles (UAVs) by implementing tracking algorithms such as DeepSort with custom-trained YoloV7 models.
- Used frameworks and libraries such as OpenCV, TensorRT, and onnxruntime in C++ to architect and develop the system, optimizing it for deployment on Jetson devices.
- Overlooking the development of an in-house image annotation tool and curation of custom datasets for deep learning projects and managing a team of 3 members.
- Employed Python and OpenCV to develop a pipeline to perform precise frame registration, aligning UAV-captured images with satellite imagery for accurate geo-location.
- Incorporated template matching algorithms with sparse, and dense optic flow to achieve close to 90% frame registration and delivered performance on HD videos with latency below 500ms, and throughput of 25fps, on a moderate capacity GPU.
Visual Computing Lab, Indian Institute of Science (IISc)
Research Intern
May '21 - April '22
Former Research Intern under Professor Anirban Chakraborty at the Visual Computing Laboratory, operated by the Department of Computational Data Sciences, at the Indian Institute of Sciences (IISc). I worked on two projects, "Superpixel Masking and Image Inpainting with Multi Exposure Fusion" and "Source Free - Multi Label Domain Adaptation (SF-MLDA)".
Primary Tasks and Responsibilities:
1. Source Free Multi-Label Domain Adaptation - SF-MLDA- Played a key role in the development of an innovative framework for performing Source-Free Multi-Label Domain Adaptation (SF-MLDA).
- Successfully integrated a co-teaching based algorithm called Divide-Mix to handle noise in training data within the SF-MLDA framework, resulting in a 7.0% improvement in accuracy.
- Assisted in the development of two networks inspired by GANs and Autoencoders respectively for anomaly detection, localization and correction.
- Experimented with structural loss and reconstruction loss to establish a correlation on image inpainting and reconstruction.
- Enhanced the pipeline by incorporating multi-exposure fusion techniques for synthetic image regeneration, achieving an impressive 80.0 per cent overall accuracy rate.
Data Science Fellow
January '21 - April '21
I was part of the four month Machine Learning Fellowship program offered by Fellowship.ai, a sister organization of launchpad.ai. During my internship, I worked on Novel Food Type Detection a project outsourced by General Electric, with a diverse team from around the world.
Primary Tasks and Responsibilities:
1. Novel Food Type Detection- Created an end-to-end functional web application using streamlit, to perform zero-shot object detection for food in an in-oven setting.
- Established baseline results with ResNet50 and ResNet101 networks by transfer lerning and training from scratch on custom datasets.
- Leveraged a state-of-the-art Contrastive Language-Image Pre-training model - CLIP by OpenAI to achieve a final Top-1 accuracy of 97.22% and Top-3 accuracy of 100.0% on a custom dataset containing ~16 images per class.
- Developed web scrapers to generate custom datasets by scraping food blogs, and Instagram using Scrapy and Selenium. Increased the size of the dataset using data augmentation techniques.
Projects
Multi-Modal Machine Learning for Object Detection.
This work was conducted as my Undergraduate degree thesis under the guidance of Prof.Pratibha N , assistant professor at the Department of Electronics and Telecommunications Engineering at BMS Institute of Technology and Management. The main aim of this project is to demonstrate the significant improvements in Computer Vision upon integrating contextual understanding used in Natural Language Processing tasks. The Contrastive Language Image Pretraining task uses multimodal data as Image-Text pairs, in Zero-shot or Few-shot settings. This work has been accepted to the 6th International Joint Conference on Advances in Computational Intelligence (IJCACI 2022).
Source Free - Multi Label Domain Adaptation
This work introduces a novel concept of Source-Free Multi-Label Domain Adaptation (SF-MLDA) using graph convolution networks (GCN) and a Co-teaching based method to tackle the problem of noisy labels. In summary, this research work aims to improve the task of adapting a deep learning model to a new domain (distribution of data) where instances have multiple labels, in the absence of a labeled source domain to aid in the adaptation process. My work was carried out in coalition with Vikash Kumar a Masters student and his advisor Prof.Anirban Chakraborty at the Indian Institute of Science, during my Internship at the Visual Computing Lab.
Superpixel Masking and Image Inpainting with Multi Exposure Fusion
The aim of this resarch was to improve Anomaly Detection and Correction using Supermixel Masking and Inpainting, in order to improve existing methods, the network was designed by employing a mask based curriculum learning approach and incorporating multi image exposure. Two variants of the network are developed taking inspiration from Generative Adversarial Networks (GAN) and Autoencoder based architectures, and this work hopes to improve applications such as image restoration, where damaged or missing parts of an image need to be reconstructed while maintaining the overall image quality. I worked with Aditya Kumar Pal a former Masters student and his advisor Prof.Anirban Chakraborty at the Indian Institute of Science, during my Internship at the Visual Computing Lab.
Dry Waste Classification Using Machine Learning and IoT
The primary objective of this project was to develop a simple yet efficient machine learning powered device to perform waste classification on organic and inorganic waste. Additionally, a paper about the work was submitted to the 2nd International Conference On Intelligent Engineering And Management, 2021. The project also aquired small scale funding of INR 20,000 to develop a fully functional prototype to scale. This project was carried out under Dr. Mallikarjuna Gowda C.P, Associate professor at the Department of Electronics and Telecommunications Engineering at BMS Institute of Technology and Management.