|
09:00 - 09:30
|
Manuscript ID. 0909
Paper No. 2022-SUN-S1005-I001
|
Invited Speaker:
Richard Hu
|
Fully automated solution for metalenses/metasurfaces with inverse design capability
Richard Hu, Synopsys (Taiwan)
Metalenses are a promising new optical technology with applications in AR/VR, biomedical imaging, and wearable consumer electronics. Owing to their planar profiles and versatility, they allow for a dramatic miniaturization of existing optical systems while realizing complex new optical functions. MetaOptic Designer, a fully automated design tool has been developed using inverse design techniques adapted for systems containing cascaded metasurfaces with arbitrary configurations of parameterized meta-atoms. The optimized layout is obtained automatically based on specified target functions. The performance of the optimized metalens system can then be validated by different simulation approaches. Several design examples will be presented to demonstrate the capability and usability of this powerful design flow.
 Preview abstract
|
|
|
09:30 - 10:00
|
Manuscript ID. 0911
Paper No. 2022-SUN-S1005-I002
|
Invited Speaker:
Chuan-Chung Chang
|
Tradeoff and system consideration for AR optical engine
Chuan-Chung Chang, Coretronic Corporation (Taiwan)
Comparison from technical to performance for different optical configuration, micro display panel and optical combiner will be given during the talks, which focus on AR optical engine to glasses especially. Several AR HMD and Glasses will be illustrated as example, and newest process in industry will be included also.
Preview abstract
|
|
|
10:00 - 10:15
|
Manuscript ID. 0877
Paper No. 2022-SUN-S1005-O001
|
| Jian-Yu Chen
|
Gaussian Belief Propagation for Bundle Adjustment with Neural Network
Jian-Yu Chen, Yu-Siang Feng, Han-Chun Wang, Chih-Wei Huang, National Central University (Taiwan); Jann-Long Cheng, National Taiwan Normal University (Taiwan)
Bundle adjustment (BA) is the major optimization step which simultaneously refines 3D coordinates and accounts for a large portion of execution time in visual simultaneous localization and mapping (SLAM). Recent solutions adopting iterative and originally slow Gaussian belief propagation (GBP) show its potential to be fast and accurate on emerging computation platforms. We propose a novel GBP-Learning architecture with deep neural networks to avoid the dependency of an arbitrary damping factor for GBP to be stabilized. Moreover, compared with standard GBP, the learning-based approach achieves the same level of accuracy while running 17.7 times faster under GPU acceleration.
Preview abstract
|
|
|
10:15 - 10:30
|
Manuscript ID. 0876
Paper No. 2022-SUN-S1005-O002
|
| Xi-Wen Chen
|
MR Application Accelerated by A Client-Server Architecture
Xi-Wen Chen, Jian-Yu Chen, Chih-Wei Huang, National Central University (Taiwan); Jann-Long Cheng, National Taiwan Normal University (Taiwan)
A client-server architecture for mixed-reality communication between the real world and a virtual object is proposed. We demonstrate the framework combined with text spotting model on HoloLens 2 so that the devices can receive the surrounding information to improve user experiments. Finally, our experiment results show that the client-server framework can provide on average 0.77 seconds of computational time per frame, which is not only 11.8 times faster than the client-only framework but also achieves near real-time computation. Moreover, the hardware usage of the client-side can be reduced to half of the client-only framework.
Preview abstract
|
|