![[2602.21341] Scaling View Synthesis Transformers](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.21341] Scaling View Synthesis Transformers
Summary
The paper explores scaling laws for view synthesis transformers, presenting a new architecture that outperforms previous models in Novel View Synthesis while optimizing compute efficiency.
Why It Matters
Understanding the scaling of view synthesis transformers is crucial for advancing computer vision technologies. This research provides insights into optimizing model architectures, which can lead to significant improvements in performance and resource utilization in real-world applications.
Key Takeaways
- Introduces the Scalable View Synthesis Model (SVSM) for efficient view synthesis.
- Demonstrates that encoder-decoder architectures can be compute-optimal contrary to previous beliefs.
- Achieves superior performance-compute trade-offs compared to existing models.
Computer Science > Computer Vision and Pattern Recognition arXiv:2602.21341 (cs) [Submitted on 24 Feb 2026] Title:Scaling View Synthesis Transformers Authors:Evan Kim, Hyunwoo Ryu, Thomas W. Mitchel, Vincent Sitzmann View a PDF of the paper titled Scaling View Synthesis Transformers, by Evan Kim and 3 other authors View PDF HTML (experimental) Abstract:Geometry-free view synthesis transformers have recently achieved state-of-the-art performance in Novel View Synthesis (NVS), outperforming traditional approaches that rely on explicit geometry modeling. Yet the factors governing their scaling with compute remain unclear. We present a systematic study of scaling laws for view synthesis transformers and derive design principles for training compute-optimal NVS models. Contrary to prior findings, we show that encoder-decoder architectures can be compute-optimal; we trace earlier negative results to suboptimal architectural choices and comparisons across unequal training compute budgets. Across several compute levels, we demonstrate that our encoder-decoder architecture, which we call the Scalable View Synthesis Model (SVSM), scales as effectively as decoder-only models, achieves a superior performance-compute Pareto frontier, and surpasses the previous state-of-the-art on real-world NVS benchmarks with substantially reduced training compute. Comments: Subjects: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) Cite as: arXiv:2602.21341 [cs.CV] (...
