[IMAGICA GROUP] Real-time sensor analysis with Machine Learning , AI and Computer Vision
DescriptionPresentation 1:
Presentation Title: Infiniworkflow: AI Vision Powered Imaging Platform
Presentation Description:
Introduction and live demo of Infiniworkflow showing rapid prototyping of workflows for building machine learning model, performing real-time inference and combining with image processing and computer vision algorithms - all with no coding and done with high performance GPU/Cloud backend.
Presentation 2:
Presentation Title: Low latency pose estimation with event camera for interactive VFX
Presentation Description:
When human pose estimation using a frame-based camera is performed in an online process during actual human motion, the latency of the estimation process causes a decrease in pose accuracy with real-time pose and estimation's runtime and camera framerate limit the time resolution of pose output. We will introduce pose updating using an event-based camera that can solve this latency and frame rate problem in the same workflow. This method computes differences sequentially as needed by computing sequential update vectors adapted to the motion magnitude, and achieves framerate improvement by immediate update of the estimated pose using the computed differences, and latency compensation by multi-step immediate update. We will also introduce our experiment in a real live scene. Our method was performed in a real live scene with dancers, and it shows that real-time display of visual effect is possible. Also showed that the proposed method compensates latency and improves framerate via pose evaluation.
(): Presentation Title: Infiniworkflow: AI Vision Powered Imaging Platform
Presentation Description:
Introduction and live demo of Infiniworkflow showing rapid prototyping of workflows for building machine learning model, performing real-time inference and combining with image processing and computer vision algorithms - all with no coding and done with high performance GPU/Cloud backend.
Presentation 2:
Presentation Title: Low latency pose estimation with event camera for interactive VFX
Presentation Description:
When human pose estimation using a frame-based camera is performed in an online process during actual human motion, the latency of the estimation process causes a decrease in pose accuracy with real-time pose and estimation's runtime and camera framerate limit the time resolution of pose output. We will introduce pose updating using an event-based camera that can solve this latency and frame rate problem in the same workflow. This method computes differences sequentially as needed by computing sequential update vectors adapted to the motion magnitude, and achieves framerate improvement by immediate update of the estimated pose using the computed differences, and latency compensation by multi-step immediate update. We will also introduce our experiment in a real live scene. Our method was performed in a real live scene with dancers, and it shows that real-time display of visual effect is possible. Also showed that the proposed method compensates latency and improves framerate via pose evaluation.
():
():
():
Event Type
Exhibitor Talks
TimeFriday, 6 December 20242:00pm - 4:00pm JST
LocationG409, G Block, Level 4
