Deva Ramanan
Carnegie Mellon University
Top CV ResearchersFrontier Research MapScore: 9h-index: 14 citations
Top CV Researcher — Rank #7 (top 10)
Professor, Robotics Institute
Contributions
object detection, pose estimation, tracking, dynamic 3D scene understanding
Why Selected
A highly influential vision researcher across detection, tracking, video, and 3D understanding, with both foundational and current visibility.
Score Breakdown
3
historical impact
2
recent visibility
2
current influence
2
asset availability
9
total
Frontier Research Map
Why Now
A sharp statement that next-frame prediction is too weak if the goal is structured understanding of dynamic physical scenes.
Key Ideas
- -The goal should be compositional 4D world representations rather than raw next-frame prediction.
- -Differentiable rendering and simulation are returning as central tools, now married to large-scale learning.
- -Humans and animals are the hard cases that expose whether a 3D representation is genuinely dynamic.
Open Questions
- ?What is the minimal structured representation needed for AR, robotics, and forecasting to share one backbone?
- ?Can 4D representations scale without collapsing into expensive engineering?
- ?Where should explicit structure sit relative to end-to-end generative models?
Canonical CV Leadershigh confidence
Sign in to access this content