CV

Master of Science in Computer Vision

Bachelor of Engineering in Electrical and Electronics Engineering (Honors)

Research Engineer, Graphics Rendering

• Building a CUDA-based Path Tracer in Python and Triton
• Also, exploring generalized neural radiance fields for 3D Scene Understanding

Graphics Research Intern

• Identified limitations in the existing research on modelling motion using implicit representations such as NeRF and proposed Deblur-NSFF, a method that can perform novel space-time synthesis from videos subject to motion blur or out-of-focus blur
• Achieved an improvement of 4.3%, and 12.8%, in PSNR and SSIM respectively, and a decrease of 28.12% in LPIPS value on modified NVIDIA Dynamic Scenes Dataset, and submitted our work to WACV 2024

Bachelor's Thesis (mentor- Prof. Narendra Ahuja)

• Extracted 3D models of pigs from videos using self-supervised learning, neural mesh renderer, and re-projection loss
• Synthesized a dataset that included segmentation masks to complement re-projection loss-based training algorithm
• Delivered results on pig tracking and pig 2D pose-estimation, contributing to a CVPRW 2021 ORAL paper on multi-view tracking

Research Intern (mentor- Prof. Jitendra Malik)

• Evaluated algorithms such as MeshRCNN, 3D-R2N2, Occupancy Networks, and GenRe for single-view 3D reconstruction of inanimate objects on a novel dataset - Amazon-Berkeley Objects dataset
• Executed preprocessing and benchmarking experiments on ABO dataset, which was accepted to CVPR 2022

Research Intern

• Explored Inflated 3D ConvNets and Temporal Segment Networks for the task of Fall Detection.
• Designed a compressive sensing framework to obtain compressed measurements of video sequences as spatio-temporal input.
• Implemented two reconstruction architectures using densely connected neural networks and Convolutional Neural Network (ReconNet), to retrieve original images from compressed measurements.