Amrit Paul

Profile

Hello! I'm Amrit

Currently pursuing my Masters in Professional Computer Science from Simon Fraser University, Canada.
Research enthusiast and AI developer with over 5 years of Research experience in the field of Computer Vision and Deep Learning, for various industrial applications in the areas of self-driving vehicles, adversarial machine learning, generative modeling and explainable AI. I hope to push the boundaries in research in this field by conducting meaningful and impactful research.

Current Work

[06.2022] Our latest paper, 3D Object Detection using Spatial Graph Convolutional Networks for Point Cloud Data, is soon to be published on Arxiv. The codebase can be found here.
[08.2021] A paper on Compressed Image-to-Image Translation using contrastive loss is proposed and will soon be published on Arvix. The new SOTA method for unsupervised image-to-image translation will be available soon

Research Interests

My area of work is in the field of Autonomous Systems and Robotics, specifically in the domain of Computer Vision and Deep Learning.
Some of my past and ongoing work span 2D and 3D vision, visual perception, scene understanding, image restoration/enhancement, self-supervised representation learning, data augmentation, etc.

Education

Graduate (09/2022 - Present)

Masters of Professional Computer Science (Visual Computing Specialization), Simon Fraser University, BC, Canada

Undergraduate (07/2013 - 05/2017)

B.Tech in Electronics and Communication Engineering (major) with minor in Computer Science Engineering (minor), VIT University, India
Public online courses: CS231N: Convolutional Neural Networks for Visual Recognition, CS231A: Computer Vision, From 3D Reconstruction to Recognition, and CS224N: Natural Language Processing with Deep Learning.

Experience

Cloud DevOps Intern at Samsung Electronics, Canada --- [05/2023 - 12/2023]

Topics: AWS, Kubernetes, Terraform, Docker, Ansible, Jenkins, GitHub Actions, CircleCI

Research Engineer at Siemens, India --- [09/2021 - 08/2022]

Topics: Surprise Adequacy, Generative models, Self-supervised Learning, Explainable AI
Advised by Dr. Girish Maskeri Rama

Senior Project Engineer at Wipro, India - Collaboration with Indian Institute of Science (IISc) --- [08/2017 - 09/2021]

Topics: 3D vision, Active Learning, Assisted Annotations, MLOps
Advised from IISc
- Prof. Suresh Sundaram
- Dr. KVS Hari
Advised from Wipro
- Dr. Ramachandra Budihal
- Mr. Sreesankar R

Publications

Projects

Explainable AI (XAI)
Working on explainable AI for comprehending the logical inferences of various neural networks to improve auto annotations accuracy and improving neural network robustness.
WIRIN (Wipro-IISc Research and Innovation Network) Assisted Annotation
Worked on AI-assisted neural networks to provide end-users with the provision to easily generate reliable ground truths, thereby improving the efficiency of manual annotation.
Elevated Body Temperature using Thermal Camera - COVID
EBT estimation using a thermal camera for fever detection to determine whether a person should be allowed to board the vehicle or not (COVID). Built a quantized neural network for face detection and thermal estimation using black-body reference source running on Android hardware.
Point cloud Object Detection and Semantic Segmentation
Worked on various state-of-the-art architectures such as VoxelNet, Pixor, PointRCNN, PointGNN, etc on Kitti dataset for 3D object detection and also came up with a novel architecture for 3D object detection using Graph Convolutional Neural Networks. I also worked on various point cloud semantic segmentation algorithms during this term.
WIRIN Auto Annotate Studio
WIRIN auto annotate studio for point clouds and images- AI-assisted auto annotation tool was developed using Javascript and Flask, hosted on an IISc server, which can be used for various tasks such as Object detection, Semantic Segmentation, etc.
Advanced Perception Detection system using Monocular Camera
Developed scene understanding, object detection, and lane detection system using a monocular camera in both day/night driving conditions with pixel precision for a leading Asia Pacific customer. An annotation tool was built which can automate this process, with support for auxiliary tasks.
Facial Keypoint Detection with time series based tracking
Facial and Iris Keypoints detection using IR camera in night driving conditions with pixel precision for a leading Asia Pacific customer. Built an annotation studio to automate the process for approximately 900 million images, achieved exceptional accuracy of 90%, and generated revenue for the company.

Skills

Programing Language
Python, C++, JavaScript
Machine Learning Frameworks
PyTorch, Tensorflow, Caffe
Machine Learning Libraries
OpenCV, Numpy, Scikit, Pandas, Matplotlib, Seaborn
Robotics
ROS, Gazebo
Misc.
Docker, Kubernetes, AWS, Flask