BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Asia/Tokyo
X-LIC-LOCATION:Asia/Tokyo
BEGIN:STANDARD
TZOFFSETFROM:+0900
TZOFFSETTO:+0900
TZNAME:JST
DTSTART:18871231T000000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20250110T023312Z
LOCATION:Hall B7 (1)\, B Block\, Level 7
DTSTART;TZID=Asia/Tokyo:20241204T150800
DTEND;TZID=Asia/Tokyo:20241204T151900
UID:siggraphasia_SIGGRAPH Asia 2024_sess120_papers_865@linklings.com
SUMMARY:A Plentoptic 3D Vision System
DESCRIPTION:Technical Papers\n\nAgastya Kalra, Vage Tamaazyan, Alberto Dal
 l'olio, Raghav Khanna, Tomas Gerlich, Georgia Giannopolou, Guy Stoppi, Dan
 iel Baxter, and Abhijit Ghosh (Intrinsic); Rick Szeliski (Google Research)
 ; and Kartik Venkataraman (Intrinsic)\n\nWe present a novel multi-camera, 
 multi-modal vision system designed for industrial robotics applications. T
 he system generates high-quality 3D point clouds, with a focus on improvin
 g the completeness and reducing hallucinations for collision avoidance acr
 oss various geometries, materials, and lighting conditions. Our system inc
 orporates several key advancements: (1) a modular and scalable \textbf{Ple
 noptic Stereo Vision Unit} that captures high-resolution RGB, polarization
 , and infrared (IR) data for enhanced scene understanding; (2) an \textbf{
 Auto-Calibration Routine} that enables the seamless addition and automatic
  registration of multiple stereo units, expanding the system's capabilitie
 s; (3) a \textbf{Deep Fusion Stereo Architecture} - a state-of-the-art dee
 p learning architecture trained fully on synthetic data that effectively f
 uses multi-baseline and multi-modal data for superior reconstruction accur
 acy. We demonstrate the impact of each design decision through rigorous te
 sting, showing improved performance across varying lighting, geometry, and
  material challenges. To benchmark our system, we create an extensive indu
 strial-robotics inspired dataset featuring sub-millimeter accurate ground 
 truth 3D reconstructions of scenes with challenging elements such as sunli
 ght, deep bins, transparency, reflective surfaces, and thin objects. Our s
 ystem surpasses the performance of state-of-the-art high-resolution struct
 ured light on this dataset. We also demonstrate generalization to non-robo
 tics polarization datasets. Interactive visualizations and videos are avai
 lable at \url{https://www.intrinsic.ai/publications/siggraphasia2024}.\n\n
 Registration Category: Full Access, Full Access Supporter\n\nLanguage Form
 at: English Language\n\nSession Chair: Hao (Richard) Zhang (Simon Fraser U
 niversity, Amazon)
URL:https://asia.siggraph.org/2024/program/?id=papers_865&sess=sess120
END:VEVENT
END:VCALENDAR