Have you ever examined your home security camera footage to figure out what’s been captured? At some point, you’ve probably questioned what your camera flagged — did it record the shadow of palm tree leaves swaying in the breeze, or a coyote that ran across your front lawn?
Akash Kumar has focused his doctoral research on analyzing automated video, integrating artificial intelligence (AI) to decipher, categorize and annotate captured footage, but on a much grander scale. This fall, he will graduate with a doctorate in computer science and serve as an applied scientist at Amazon, using his expertise in video understanding to develop scalable, annotation-efficient video systems for the global powerhouse.
Kumar, a native of Patnia, India, earned his bachelor’s degree from Delhi Technological University. Surprisingly, he didn’t begin in academic career with an outright interest in computer science. Kumar ranked in the Top 50 in India in robotics competitions as an undergraduate, and his experiences in automation sparked his interest in the field.
He chose to pursue a doctoral degree to dive deeper into computer science and applied to UCF, drawn to the university’s stellar reputation as a leader in computer vision research.
“The university is among the Top 10 globally in this domain, making it the ideal environment to grow as a researcher,” he says.
He thrived as a researcher for the world-renown Center for Research in Computer Vision (CRCV), guided by his advisor, Associate Professor Yogesh Singh Rawat.
“My research focuses on annotation-efficient dense scene understanding, essentially, figuring out ‘who’ and ‘where’ in a scene with minimal labeled data,” he says. “I was drawn to this because real-world deployment of AI systems often depends on models that can adapt to new environments without requiring extensive manual labeling. My goal is to develop methods that scale and work robustly in practical, dynamic settings.”
His work at CRCV helped land him internships at the company that would eventually hire him upon graduation, Amazon. As an intern, he drew upon his expertise in machine learning to support their video operations.
During his first internship, Kumar designed an object detection system that utilizes an open-vocabulary framework, or a flexible system not bound by pre-defined categories. Using video and motion cues, his system greatly improved accuracy over its existing processes.
Last summer, he developed a framework for on-device instance segmentation, a technique that identifies and categorizes multiple similar objects, improving Amazon’s labeling and computing efficiency by 80%.
As he looks back on his time at UCF, his milestones as a doctoral student stand out as the most memorable, especially his first time presenting his research.
“After a year of work, sharing a novel approach with the community was deeply rewarding,” he says. “It clarified the type of research I wanted to pursue, bridging innovation between academia and industry.”
He adds that because joint space and time understanding in computer vision is still a burgeoning discipline, he’s most proud of publishing research in such a specialized field. Kumar says only a handful of papers are published each year in this challenging area.
“With so few prior works, extracting insights for my own problem often meant starting from scratch,” he says. “In several cases, I was the first to bring together concepts in a meaningful way. Despite numerous rejections, I published 12 papers, which strengthened my resilience and deepened my expertise.”
He’s grateful that his graduate training led to these valuable experiences, and is confident that having a doctoral degree will pave the way for even greater opportunities.
“A doctorate opens doors to senior-level positions across top companies, often enabling mid-career-level roles early due to the depth of research experience,” Kumar says. “It also allows for meaningful collaboration and conversations at premier conferences, where our research aligns with real-world industry problems.”
- Written by Bel Huston