SparseStreet: Sparse Gaussian Splatting for Real-Time Street Scene Simulation Conference: International Conference on Multimedia Retrieval; June 16–19, 2026; Amsterdam, Netherlands International Conference on Multimedia Retrieval (ICMR ’26), June 16–19, 2026, Amsterdam, Netherlands DOI: 10.1145/3805622.3810887 ISBN: 979-8-4007-2617-0/2026/06 CCS: Computing methodologies Reconstruction CCS: Computing methodologies Rendering CCS: Computing methodologies Supervised learning
Qingpo Wuwu, Xiaobao Wei, Peng Chen, Nan Huang, Zhongyu Zhao, Hao Wang, Ming Lu, Ningning Ma, Shanghang Zhang · 2026
Read SparseStreet: Sparse Gaussian Splatting for Real-Time Street Scene Simulation Conference: International Conference on Multimedia Retrieval; June 16–19, 2026; Amsterdam, Netherlands International Conference on Multimedia Retrieval (ICMR ’26), June 16–19, 2026, Amsterdam, Netherlands DOI: 10.1145/3805622.3810887 ISBN: 979-8-4007-2617-0/2026/06 CCS: Computing methodologies Reconstruction CCS: Computing methodologies Rendering CCS: Computing methodologies Supervised learning by Qingpo Wuwu, Xiaobao Wei, Peng Chen, Nan Huang, Zhongyu Zhao, Hao Wang, Ming Lu, Ningning Ma, Shanghang Zhang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-03909.txt.
Description
Wuwu and co-authors compress street scene reconstructions while keeping dynamic objects crisp. 3D Gaussian Splatting rebuilds street scenes well. Fine detail demands huge numbers of primitives, though. Storage balloons and rendering slows. Vehicles and pedestrians need high fidelity for temporal consistency. Static backgrounds, by contrast, hold heavy redundancy.
SparseStreet exploits that split. A node-based learnable pruning step removes low-contribution primitives. Visually critical regions are protected during the cut. Once the scene settles, a second pass compresses the background further. Geometry and appearance of dynamic objects survive the squeeze. On Waymo and nuScenes data, compression reaches up to 80 percent. Quality loss stays minimal through the cut. Real-time street simulation becomes feasible on leaner hardware. Simulation teams gain fidelity without the storage bill.
Catalog record
| CREATOR | Qingpo Wuwu, Xiaobao Wei, Peng Chen, Nan Huang, Zhongyu Zhao, Hao Wang, Ming Lu, Ningning Ma, Shanghang Zhang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.03909