MREENAV S DEKA
professional bio
MREENAV S DEKA
COMPUTER VISION | MACHINE LEARNING ENGINEER | PART-TIME MUSICIAN
WORK
I love Computer Vision and exploring new technologies.
- computer vision
- machine learning
- neural scene representations
- object detection
RESEARCH ASSISTANT
Jul 2023-Present
Researching on applying domain adaption to object detection models. My responsibilities include reviewing literature, implementing papers, benchmarking models, and presenting results.
TECH SPECIALIST 1 - MACHINE LEARNING ENGINEER
Nov 2022-Mar 2023
Developed multiple machine learning pipelines from scratch. Researched on densely placed small object detection (around 5000 detections per image) and achieved a mAP of 90%. Deployed solution using Gradio. Implemented pipelines for object classification and auto-labelling.
SOFTWARE ENGINEER
Aug 2019-Mar 2021
Development and maintenance of the CGI Trade360 trade finance platform. Worked on the development of APIs, creation of user interfaces, and defect fixing as per client requirements. Lead multiple feature developments. Worked on asynchronous background processes that handle the validity of transactions.
Data Science Intern
Jun 2017-Jul 2017
Created the initial data pipeline which is used to scrape, parse, and store data from online websites dealing with innovations in Biotechnology so that it can be used to inform business decisions. Worked under the mentorship of Mr. Divya Shakti, who is the co-founder of BiosectRx.
RESEARCH WORK
Joint Camera Refinement and Transient Object Removal for Street View Synthesis
Arxiv 2023
Mreenav Shyam Deka*, Lu Sang*, Daniel Cremers
Our method generates high-fidelity street view imagery while autonomously managing dynamic moving objects, eliminating the need for manual annotations. Additionally, we simultaneously refine the initial camera poses to enhance the quality of the renderings.
Weakly Supervised CAD Retrieval from a Single RGB Image
2023
Mreenav Shyam Deka, Angela Dai
I developed a weakly supervised method that uses a perceptual loss function to create an embedding space between RGB images and CAD meshes. By detecting objects in an image, we can retrieve the corresponding CAD object to represent the scene.
MUSIC
I love writing and producing music, and occasionally performing live.
Waves
Featuring Mrinmoy Brahma
Represents the human emotion of clinging on to the past.
Daft Lebowski
Featuring Nandini Sharma
An experiment with the vocoder feat. Non's voice.
Stranded
With Theps, Abhigyan, and Joel
First studio-recording, a collaboration about moving out of your house for the first time.
BACKGROUND
QUALIFICATIONS / CERTIFICATIONS
Technische Universität München
Oct 2021 - Current
MSc Data Engineering and Analytics
NIIT Stackroute Koramangala
Sep 2019-Dec 2019
14 week full stack development bootcamp. Java Immersive.
National Institute of Technology Silchar
Jul 2015 - Jun 2019
Bachelor of Technology in Computer Science and Engineering.