Cloud-Based Dynamic Streaming and Loading of 3D Scene
Download as PPTX, PDF0 likes77 views
The document discusses a cloud-based dynamic streaming approach for visualizing large 3D urban scenes, allowing scalable loading of geometry on-demand to improve performance and memory usage. It highlights the advantages of this method over traditional loading techniques, particularly in applications such as urban planning and virtual reality simulations. The proposed system leverages client-server communication and stores geometry in cloud formats, demonstrating efficient handling of complex datasets.
![Introduction
• Visualization of very large scene using out-of-core approach
– Client application does not have enough resource to load all data
• Aim for scalability
• Cloud-based geometry loading approach
– Geometry is stored in cloud
– Geometry is streamed and loaded on-demand
• Other approaches:
– Cloud rendering : Deng et al. [1]
– Out-of-core CPU to GPU : Crassin et al. [3]](https://image.slidesharecdn.com/cyberworlds2018presentationntuthemenovideo-190221053532/85/Cloud-Based-Dynamic-Streaming-and-Loading-of-3D-Scene-3-320.jpg)
![Related Work
• Local out-of-core storage approach
– Room subdivision : [4]
– Preprocessed assets such as in Unity and Unreal Engine [5-10]
• Adaptive loading
– based on current and previous state : He [19]
– based on geometry importance : Tian and AlREgib [11]
– based on viewing frustum : Zhi et al. [20]](https://image.slidesharecdn.com/cyberworlds2018presentationntuthemenovideo-190221053532/85/Cloud-Based-Dynamic-Streaming-and-Loading-of-3D-Scene-7-320.jpg)
![Related Work
• Stream precomputed illumination and progressive mesh :
Pacanowski et al. [12]
• Regular scene subdivision : He [11] and Zhi et al. [15]
• Optimize texture streaming : Eu et al. [14] and Englert et al. [15]
• Attach geometry to other file format : Concolato et al. [16]
• Distributed P2P streaming: Wang et al. [17] and Jie et al. [18]](https://image.slidesharecdn.com/cyberworlds2018presentationntuthemenovideo-190221053532/85/Cloud-Based-Dynamic-Streaming-and-Loading-of-3D-Scene-8-320.jpg)
Cloud-Based Dynamic Streaming and Loading of 3D Scene
- 1. Cloud-Based Dynamic Streaming and Loading of 3D Scene Budianto Tandianus1 , Hock Soon Seah1 Tuan Dat Vu2, Anh Tu Phan2 1. Multi-plAtform Game Innovation Centre, Nanyang Technological University, Singapore 2. School of Information and Communication Technology, Hanoi University of Science and Technology, Vietnam
- 3. Introduction • Visualization of very large scene using out-of-core approach – Client application does not have enough resource to load all data • Aim for scalability • Cloud-based geometry loading approach – Geometry is stored in cloud – Geometry is streamed and loaded on-demand • Other approaches: – Cloud rendering : Deng et al. [1] – Out-of-core CPU to GPU : Crassin et al. [3]
- 4. Introduction • Basic client-server communication – Client sends request to server to query for nearby buildings • Information sent: position, location – Server returns geometry • Client receives the geometry and import it to the scene – Client implemented in Unity – Geometry in CityGML format • Client discards geometry whose distance is greater than specified radius
- 5. Contribution • Scalable cloud-based urban simulation approach – Can be used in various applications such as visualization and games – Designed to handle non-urban objects (e.g. vehicles and humans), not only urban objects (e.g. buildings) • We show the advantage of our approach compared to traditional loading (in term of memory usage and rendering performance) – Able to handle very large scene (more than 1 million polygons) without large drop in system performance.
- 6. Application • Urban planning and development using Virtual Reality – Visualize semantic and dynamic information – Explore virtual world • Seamless virtual simulation of an entire metropolis • Beneficial for applications the require intuitive path planning and interaction: – Building and urban design visualization – Evacuation planning – Semantic crowd simulation – Computer games
- 7. Related Work • Local out-of-core storage approach – Room subdivision : [4] – Preprocessed assets such as in Unity and Unreal Engine [5-10] • Adaptive loading – based on current and previous state : He [19] – based on geometry importance : Tian and AlREgib [11] – based on viewing frustum : Zhi et al. [20]
- 8. Related Work • Stream precomputed illumination and progressive mesh : Pacanowski et al. [12] • Regular scene subdivision : He [11] and Zhi et al. [15] • Optimize texture streaming : Eu et al. [14] and Englert et al. [15] • Attach geometry to other file format : Concolato et al. [16] • Distributed P2P streaming: Wang et al. [17] and Jie et al. [18]
- 9. System Design • Cloud Storage: store CityGML models • Database: indexing and retrieving buildings – Use PostgreSQL – Future plan: non-urban objects such as animals and NPCs. More flexibility in the future such as splitting objects. • API Service: PHP and Yii2 • Internal Communication: RESTFUL APIs – Send location to server and sever returns a set of unique IDs of 3D objects within predefined radius – IDs are used to retrieve geometry from cloud storage • Renderer: Unity – Use modified Jaeniche’s script for loading CityGML data
- 10. System Design A cloud based system to streamline data management Media Transcoding & Streaming Cross Platform Accessibility Expandability via Virtual User Flexible User Role Management Efficient Workflow Management Secure File Sharing Powerful Asset Management & Tracking Provide a more efficient and collaborative platform
- 11. Experiment Setup • Two datasets: – Berlin : 17,122 polygons, 954 files, 38.2 MB – Amsterdam : 1,221,379 polygons, 90,260 files, 906 MB • Workstation specification: – CPU : Intel(R) Xeon(R) CPU E5-2609 0 @ 2.4GHz 16 GB RAM – GPU : NVIDIA Quadro M4000 16GB • Traditional Loading vs Dynamic Streaming – Compared memory usage – Frames-per-second – Rendering time per-frame Berlin Dataset Amsterdam Dataset
- 12. Result Berlin dataset Traditional average: 43.78 MB Dynamic average: 1.81 MB Amsterdam dataset Dynamic average: 1.62 MB
- 13. Result Berlin dataset Traditional average: 9.16 fps Dynamic average: 30.54 fps Amsterdam dataset Dynamic average: 29.99 fps
- 14. Result Berlin dataset Traditional average: 103.16 ms Dynamic average: 32.07 ms Amsterdam dataset Dynamic average: 32.83 ms
- 15. Result Traditional Loading Dynamic Streaming
- 17. Conclusion • Promising improvement using dynamic streaming • Low memory usage • High FPS
- 18. Future Work • Asset allocation and deallocation based on prediction on player’s action history – Camera speed and orientation • Storing precomputed illumination on hierarchical grid • Perceptual-based optimization – Progressive mesh refinement – Texture streaming • Large scene simulation with path planning • Deploy to standalone VR device – Oculus Go – HTC Vive Focus
- 19. Acknowledgement • This research is supported by the National Research Foundation, Prime Minister’s Office, Singapore under its IDM Futures Funding Initiative.
- 20. The End