Achievements
DAAD PhD Research Grant
Technical University Berlin, Germany (August 2017 - March 2022)
This prestigious grant, offered by the German Academic Exchange Service (DAAD), supported full-time PhD research, allowing advanced study in computer vision and machine learning at a leading German institution, fostering international academic collaboration.
Best Student Award
IIT Roorkee, India (Sept. 2016)
Recognized for outstanding academic performance in the M.Tech. Engg./M.Arch. programs (2014-2016), this award highlights excellence in coursework and research, reflecting dedication to engineering innovation.
Dr. Jai Krishna Medal
IIT Roorkee, India (Sept. 2016)
Awarded for achieving the highest CGPA across all M.Tech. courses, this medal honors academic rigor and is named after Dr. Jai Krishna, a pioneer in earthquake engineering, symbolizing excellence in technical education.
Academic Prize
IIT Roorkee, India (Sept. 2016)
Granted for the highest CGPA in the M.Tech. (Electrical Engineering) final year, this prize underscores exceptional mastery of electrical engineering concepts and their practical applications.
DAAD IIT Master Sandwich Scholarship
Technical University Berlin, Germany (Sept 2015 - March 2016)
This DAAD-funded scholarship enabled a final-year master's student to conduct full-time dissertation research in Germany, enhancing skills in 3D reconstruction and fostering global research networks.
Doctoral Thesis
Deep Learning Mesh Parameterization of 3D Shapes: For 3D Reconstruction, Shape Generation, Noise Filtering, and Mobile Rendering
Deep learning has made remarkable progress in extracting meaningful information from visual sensory 2D data. The regularized grid representation of the 2D image data makes it convenient to apply efficient convolutional kernels. This thesis employs 2D convolutional kernels for 3D surface meshes using mesh parameterization, enabling efficient networks that can be inferenced on low-compute devices.
Research Publications
IMD-Net: a Deep Learning-based Icosahedral Mesh Denoising Network
We propose a novel denoising technique, the icosahedral mesh denoising network (IMD-Net) for closed genus-0 meshes. Enabled by gauge equivariant convolutional layers arranged in a residual U-net, IMD-Net denoises the remeshing invariant to global mesh transformations.
GenIcoNet: Generative Icosahedral Mesh Convolutional Network
We propose a generative icosahedral mesh convolutional network (GenIcoNet) that learns data distribution of surface meshes. Our end-to-end trainable network learns semantic representations using 2D convolutional filters on regularized icosahedral meshes.
Learning to Reconstruct Symmetric Shapes using Planar Parameterization of 3D Surface
We reconstruct 3D shapes from images using parameterized representation. We perform iterative parameterization to obtain planar representation encoded with surface information to generate 2D geometry images, which can be learned using traditional deep neural networks.
Improving 3D Face Geometry by Adapting Reconstruction From Stereo Image Pair to Generic Morphable Model
We remedy stereo reconstruction deficits by combining 3D stereo reconstruction with a generic Morphable Model, allowing to prefer information from single image reconstruction whenever stereo reconstruction shows untypical deviations from expected 3D features.
Master Thesis
Using Morphable Face Model to Improve Stereo Reconstruction and Visualising the Model on a Smartphone
This work remedies stereo reconstruction deficits by combining 3D stereo reconstruction with a generic Morphable Face Model. The fusion outcome results in geometrically more accurate face reconstruction, visualized on a smartphone using cardboard.
Publications from Accomplished Projects
Calibration and Registration Method for Tomography-Based Laser-Guided Surgical Interventions using a 4-DOF Navigation Robot
A method for co-registering pre-surgically acquired tomographic data with a patient at time of surgical intervention. The achievable accuracy of the method is sufficient for periradicular therapy.
Laser-Guided CT Intervention using Flexible Laser Bow
We introduce an image-based laser guided CT intervention with a flexible laser bow system. The proposed approach allows movability to the laser bow system without compromising on accuracy.
A Power Efficiency Enhancements of a Multi-Bit Accelerator for Memory Prohibitive Deep Neural Networks
We propose a power efficient multi-bit neural network accelerator, employing the technique of truncating the partial sum (PSum) results from the previous layer before feeding it into the next layer.
Patient Motion Compensation for Photogrammetric Registration
We present two approaches for motion compensation: disparity shift compensation, and moving cameras compensation - both capable of achieving patient registration qualitatively equivalent to motion-free registration.
Passive Classification of Source Printer using Text-line-level Geometric Distortion Signatures from Scanned Images of Printed Documents
Ability to easily and reliably identify source printer of a printed document can help reduce forgeries on legal tenders. We propose a method using text-line-level geometric distortion signatures from scanned images.
An Enhanced Statistical Approach for Median Filtering Detection using Difference Image
In image forensics, detection of image forgeries involving non-linear manipulations have received great interest. We propose an enhanced statistical approach for detecting median filtering operations.
Invited Talks
Intro to Deep Volumetric, Point Cloud & Mesh Learning
An introductory session on deep learning approaches for volumetric data, point clouds, and mesh representations, covering fundamental concepts and recent advancements in 3D vision.
Deep Learning Explicit 3D Representations
Exploring deep learning techniques for explicit 3D representations, focusing on efficient modeling and processing of 3D geometric data for computer vision applications.
3D Computer Vision and Beyond
A comprehensive overview of 3D computer vision techniques, current challenges, and future directions, including applications in autonomous systems, robotics, and immersive technologies.
AICTE Faculty Development Programme
A session on Introduction to 3D computer vision and 3D medical data registration for faculty members working in the domain of healthcare.
3D Medical Data Registration
Online guest lecture for Healthcare Technology students on modern approaches for 3D medical data registration and its practical applications.
DAAD Scholarship Meeting
Invited talk on immersive virtual reality experiences during the DAAD Scholarship Meeting, highlighting VR applications in science communication.
Talk at my School
Interactive session with class 12 students focused on real educational experiences, common mistakes, and strategies to build a resilient academic journey.
Technical Contributions
C++ Implementation of Iterative Parameterization
Contributed to the Computational Geometry Algorithms Library (CGAL) by implementing an iterative approach for authalic parameterization of surface meshes. This enhancement addresses the limitation of existing methods in handling surface meshes with large Gaussian curvature, providing near-optimal solutions for reducing distortions in planar parameterization of triangulated surface meshes.
3D Reconstruction Android Application
Developed an Android application that implements the research paper "Learning to Reconstruct Symmetric Shapes using Planar Parameterization of 3D Surface". The app performs real-time 3D reconstruction from single images using deep learning models, specifically trained for airplanes and cars. Users can capture images, generate 3D models, render them on mobile devices, and export in various 3D formats.
VR Mobile Applications
CV VR
A virtual reality application premiered at Long Night of Science 2019 in Berlin, Germany. This application showcases projects from the computer vision and remote sensing group at Technical University Berlin, featuring 3D Toronto city model reconstruction, dinosaur skeleton point cloud visualization, and train station 3D reconstruction.
GRSS VR
An educational VR application that illustrates the work of the seven technical committees of the Geoscience and Remote Sensing Society (IEEE GRSS). The app includes seven immersive scenes, one for each committee, plus an introduction scene to help users get acquainted with cardboard operations. Compatible with Google Cardboard V2.
Birthday VR
The first-ever human model-based Virtual Reality application on Google Play Store. This innovative app features three interactive gift boxes enclosed with human models that trigger animations and audio when gazed at for more than 2 seconds. The application required advanced C# scripting and custom humanoid model animations, with easy switching between VR and full-screen modes.