Ankita Christine Victor

I currently work as a Software Engineer at Microsoft where I write code for distributed data storage that supports big data analytics worksloads with high throughput and low latency requirements.

I graduated from the International Institute of Information Technology Bangalore in July 2019, with a Bachelor's and a Master's degree in Information Technology and a specialization in Data Science. At university, I was the editor of the college magazine, an executive organizer for TEDxIIITBangalore, graphic designer for the Branding and Outreach team, and on the Dean's Merit List for academic achievement. My master's thesis on image to 3D scene construction was supervised by Prof. Jaya Sreevalsan Nair and mentored by Prof. T. K. Srikanth. .

Email  /  CV  /  Twitter  /  Github

profile photo

My interests lie primarily in the nexus of machine learning, vision, and computer graphics. I am deeply intrigued by the applications of machine learning in modeling and the generation of virtual environments. My graduate research looked at image to 3D synthesis, specifically the design of a system that can automate the process of modeling 3D environments for autonomous vehicle simulation

When I get the time, I am still trying to write the perfect rendering engine in OpenGL, and am fascinated by geometry, geometric modeling and polygonal mesh processing.

Scene Editing Using Synthesis of Three-Dimensional Virtual Worlds From Monocular Images of Urban Road Traffic Scenes
Ankita Christine Victor, Jaya Sreevalsan Nair
The 16th ACM SIGGRAPH European Conference on Visual Media Production, 2019

Proposed a workflow to construct a skeleton 3D scene from a monocular image which can either be rendered as-is or edited to add more detail. The workflow uses a convolutional neural network (CNN) to estimate depth for a semantically segmented image and computes a matrix to correct the effects of perspective projection using direct linear transform (DLT).

Synthesis of Three-Dimensional Virtual Worlds From Monocular Images of Urban Road Traffic Scenes
Ankita Christine Victor, Jaya Sreevalsan Nair
Graduate Thesis

Researched potential solutions to automate the process of modeling of urban traffic scenes. Proposed a workflow that given a monocular image, uses a CNN to generate a dense depth map, semantic segmentation to understand the scene, inverse projection to correct perspective distortion in the image, and position comparisons to correct errors in depth. A rendering engine loads and display 3D models belonging to a particular semantic class at the computed 3D position.

Learning Lip Sync from Audio
Ankita Christine Victor

Generated natural looking 2D speech animation that synchronizes with audio and is then composited onto a target video clip. Used a recurrent neural network with LSTM units to learns the mapping from raw audio features to mouth shapes. Given the mouth shape at each time instant, texture is synthesized using a conditional adversarial network and composited onto a target frame so that the target's mouth appears to match an input audio track.

CNN Based Monocular Image Augmentation
Ankita Christine Victor, Jaya Sreevalsan Nair

Explored rendering an image as a point cloud by using a convolutional neural network to estimate per pixel depth. The image was rendered as a collection of textured sprites and augmented with 3D mesh models.

Billboard Based Monocular Augmentation
Ankita Christine Victor, TK Srikanth

Augmented monocular video frames with realistic occlusion. Objects of interest in the foreground were billboarded and each frame was rendered as a set of billboards at different positions along the Z-axis. New mesh models were then inserted and animated as such.

Teaching Experience
Graduate Student Instructor, Computer Graphics
Undergraduate Course, Spring 2019

Graduate Student Instructor, Introduction to Computer Graphics
Graduate Elective, Fall 2017

Website template taken from Jon Barron