Spatial AI Lab – Zurich: Advancing AI and Computer Vision Research

Spatial AI Lab – Zurich: Advancing the Frontiers of Artificial Intelligence and Computer Vision

This document outlines the research focus and publications of the Spatial AI Lab at Microsoft Research in Zurich. The lab is dedicated to pushing the boundaries of artificial intelligence, with a particular emphasis on computer vision, spatial understanding, and their applications in areas like robotics and autonomous systems.

Research Areas:

The Spatial AI Lab's research spans several key areas within AI and computer vision:

• Intelligence:
  • Artificial intelligence
  • Audio & acoustics
  • Computer vision
  • Graphics & multimedia
  • Human-computer interaction
  • Human language technologies
  • Search & information retrieval
• Systems:
  • Data platforms and analytics
  • Hardware & devices
  • Programming languages & software engineering
  • Quantum computing
  • Security, privacy & cryptography
  • Systems & networking
• Theory:
  • Algorithms
  • Mathematics
• Other Sciences:
  • Ecology & environment
  • Economics
  • Medical, health & genomics
  • Social sciences
  • Technology for emerging markets

Key Research Themes and Publications:

The lab's work is characterized by a strong focus on practical applications and theoretical advancements. Several publications highlight their contributions:

• Understanding the Limitations of CNN-based Absolute Camera Pose Regression: This research delves into the challenges and limitations of using Convolutional Neural Networks (CNNs) for estimating a camera's absolute pose, a critical task in computer vision and robotics.
• Project AutoVision: Localization and 3D Scene Perception for an Autonomous Vehicle with a Multi-Camera System: This project focuses on enabling autonomous vehicles to understand their surroundings by accurately localizing themselves and perceiving the 3D structure of the environment using multiple cameras.
• Approximate Convex Decomposition and Transfer for Animated Meshes: This work addresses the problem of decomposing complex 3D animated meshes into simpler convex shapes, which is important for efficient rendering and simulation.
• DGC-Net: Dense Geometric Correspondence Network: This research introduces a novel network for finding dense geometric correspondences between 3D shapes, crucial for tasks like object recognition and pose estimation.
• Towards Robust Visual Odometry with a Multi-Camera System: This paper explores methods for achieving robust visual odometry (estimating a camera's motion from video) using multiple cameras, enhancing the reliability of navigation systems.
• Learning to Fuse Proposals from Multiple Scanline Optimizations in Semi-Global Matching: This work focuses on improving stereo matching algorithms by learning to effectively combine information from different optimization strategies.
• VSO: Visual Semantic Odometry: This research integrates semantic information into visual odometry to improve its accuracy and robustness, particularly in complex environments.
• A Dataset of Flash and Ambient Illumination Pairs from the Crowd: This publication introduces a new dataset designed to help researchers develop and evaluate algorithms for handling varying illumination conditions.
• Semantic Match Consistency for Long-Term Visual Localization: This work proposes a method for robustly localizing a camera in a large-scale environment over extended periods by leveraging semantic information.
• Learning Priors for Semantic 3D Reconstruction: This research explores how to learn prior knowledge to improve the process of reconstructing 3D scenes with semantic understanding.

Leadership and Contact Information:

The Spatial AI Lab is led by Marc Pollefeys, Partner Director of Science. The lab is located at Seestrasse 356, Zurich 8038, Switzerland.

Social Media and Engagement:

The lab actively engages with the research community through various platforms:

• Social Media: Facebook, X (formerly Twitter), LinkedIn, Instagram, YouTube.
• Community Engagement: Microsoft Research blog, Microsoft Research Forum, podcasts, and newsletters.

Further Information:

• Lab Overview: Link to Overview
• People: Link to People
• Publications: Link to Publications

This summary provides a glimpse into the cutting-edge research conducted at the Spatial AI Lab in Zurich, contributing significantly to the fields of artificial intelligence and computer vision.