Pushmeet Kohli

About

Pushmeet Kohli is a leading researcher in interactive computer vision and machine learning, best known for pioneering work that bridges human input with automated scene understanding. His key research areas include interactive 3D segmentation, Gaussian process optimization, and user-centric learning systems. Kohli’s most impactful contribution is the development of SemanticPaint, an interactive 3D scene understanding system that allows users to simultaneously scan and segment their environment by simply touching objects. This work, which has garnered over 160 citations across its two seminal papers, revolutionized how machines can learn from real-time human interaction. He has also made significant advances in batched Gaussian process bandit optimization using determinantal point processes, a method critical for efficient hyper-parameter tuning in machine learning models. Beyond these technical achievements, Kohli has been a strong advocate for user-centric evaluation in interactive segmentation, arguing that systems must adapt to diverse users rather than treating all interactions as identical. His research has profoundly influenced how modern AI systems incorporate human guidance, making him a key figure in the development of practical, interactive AI tools.

Research Focus

Key Achievements

Top Papers

1. 1
   SemanticPaint

   83 citations · 2015
2. 2
   SemanticPaint: Interactive 3D Labeling and Learning at your Fingertips

   77 citations · 2015
3. 3
   Batched Gaussian Process Bandit Optimization via Determinantal Point\n Processes

   39 citations · 2016
4. 4
   User-Centric Learning and Evaluation of Interactive Segmentation Systems

   26 citations · 2012
5. 5
   Learning an interactive segmentation system

   24 citations · 2010
6. 6
   Adapting Interaction Environments to Diverse Users through Online Action Set Selection

   9 citations · 2014
7. 7
   Learning an Interactive Segmentation System

   3 citations · 2009

Key Collaborators

Contact & Links