Details
| Originalsprache | Englisch |
|---|---|
| Titel des Sammelwerks | MMVE 2024 - Proceedings of the 2024 16th International Workshop on Immersive Mixed and Virtual Environment Systems |
| Seiten | 78 - 84 |
| Seitenumfang | 7 |
| ISBN (elektronisch) | 9798400706189 |
| Publikationsstatus | Veröffentlicht - 15 Apr. 2024 |
Abstract
Progressive coding is a valuable technique for networked immersive media. As users approach objects in an immersive environment, progressive coding enables a gradual improvement of content quality. This effectively reduces bandwidth consumption compared to non-progressive methods that require to fully exchange a content representation by an independent, new representation. In this work, we introduce an approach to progressively code point cloud attributes in a learned manner by compressing quantization residuals of each preceding representation through a learned, lightweight transformation in the entropy bottleneck. This allows to progressively reduce quantization errors using a single model in an end-to-end learning manner given the quantization residuals. In contrast to the state of the art that conditions the compression on a fixed rate-distortion, i.e. it requires an ensemble of models to build an adaptive streaming system, our approach requires only a single model during compression and decompression. We present preliminary results of our method, showing bandwidth savings for the scenario of a user approaching an object and gradually transitioning from low to high quality representations.
ASJC Scopus Sachgebiete
- Informatik (insg.)
- Computergrafik und computergestütztes Design
- Informatik (insg.)
- Mensch-Maschine-Interaktion
- Informatik (insg.)
- Software
Zitieren
- Standard
- Harvard
- Apa
- Vancouver
- BibTex
- RIS
MMVE 2024 - Proceedings of the 2024 16th International Workshop on Immersive Mixed and Virtual Environment Systems. 2024. S. 78 - 84.
Publikation: Beitrag in Buch/Bericht/Sammelwerk/Konferenzband › Aufsatz in Konferenzband › Forschung › Peer-Review
}
TY - GEN
T1 - Progressive Coding for Deep Learning based Point Cloud Attribute Compression
AU - Rudolph, Michael
AU - Riemenschneider, Aron
AU - Rizk, Amr
N1 - Publisher Copyright: © 2024 ACM.
PY - 2024/4/15
Y1 - 2024/4/15
N2 - Progressive coding is a valuable technique for networked immersive media. As users approach objects in an immersive environment, progressive coding enables a gradual improvement of content quality. This effectively reduces bandwidth consumption compared to non-progressive methods that require to fully exchange a content representation by an independent, new representation. In this work, we introduce an approach to progressively code point cloud attributes in a learned manner by compressing quantization residuals of each preceding representation through a learned, lightweight transformation in the entropy bottleneck. This allows to progressively reduce quantization errors using a single model in an end-to-end learning manner given the quantization residuals. In contrast to the state of the art that conditions the compression on a fixed rate-distortion, i.e. it requires an ensemble of models to build an adaptive streaming system, our approach requires only a single model during compression and decompression. We present preliminary results of our method, showing bandwidth savings for the scenario of a user approaching an object and gradually transitioning from low to high quality representations.
AB - Progressive coding is a valuable technique for networked immersive media. As users approach objects in an immersive environment, progressive coding enables a gradual improvement of content quality. This effectively reduces bandwidth consumption compared to non-progressive methods that require to fully exchange a content representation by an independent, new representation. In this work, we introduce an approach to progressively code point cloud attributes in a learned manner by compressing quantization residuals of each preceding representation through a learned, lightweight transformation in the entropy bottleneck. This allows to progressively reduce quantization errors using a single model in an end-to-end learning manner given the quantization residuals. In contrast to the state of the art that conditions the compression on a fixed rate-distortion, i.e. it requires an ensemble of models to build an adaptive streaming system, our approach requires only a single model during compression and decompression. We present preliminary results of our method, showing bandwidth savings for the scenario of a user approaching an object and gradually transitioning from low to high quality representations.
KW - 6DOF
KW - Adaptive Streaming
KW - Point Cloud
KW - Virtual Reality
UR - http://www.scopus.com/inward/record.url?scp=85191529111&partnerID=8YFLogxK
U2 - 10.1145/3652212.3652217
DO - 10.1145/3652212.3652217
M3 - Conference contribution
SP - 78
EP - 84
BT - MMVE 2024 - Proceedings of the 2024 16th International Workshop on Immersive Mixed and Virtual Environment Systems
ER -