通过跳过冗余梯度计算，让3D高斯训练快23%。
SkipGS: Post-Densification Backward Skipping for Efficient 3DGS Training
- 按视图动态决定是否反向传播，减少无效计算。
- 在Mip-NeRF 360上训练时间减少23.1%，后密度阶段提速42.0%。
- 无需改动渲染器或模型，可与其它优化方法叠加使用。
3D高斯点云（3DGS）通过优化数百万个各向异性高斯点实现实时新视角合成，但其训练成本仍高，尤其在后密度细化阶段，反向传播占主导。我们发现该阶段大量采样视图的损失已趋于平稳，梯度收益递减，但标准训练仍执行完整反向传播。为此提出SkipGS，引入一种视图自适应的反向传播门控机制。该方法始终进行前向传播以更新每视图损失统计，当某视图损失稳定在近期基准附近时则跳过反向传播，同时保证最低反向传播预算以维持优化稳定性。在Mip-NeRF 360数据集上，相比3DGS，SkipGS将端到端训练时间减少23.1%，其中后密度阶段时间降低42.0%，重建质量保持相当。由于仅改变反向传播时机而不修改渲染器、表示或损失函数，SkipGS具有即插即用性，可与其他效率策略协同实现累加加速。代码已开源：https://github.com/ASU-ESIC-FAN-Lab/SkipGS。
原文摘要 · Abstract (English)
3D Gaussian Splatting (3DGS) achieves real-time novel-view synthesis by optimizing millions of anisotropic Gaussians, yet its training remains expensive, with the backward pass dominating runtime in the post-densification refinement phase. We observe substantial update redundancy in this phase: many sampled views have near-plateaued losses and provide diminishing gradient benefits, but standard training still runs full backpropagation. We propose SkipGS with a novel view-adaptive backward gating mechanism for efficient post-densification training. SkipGS always performs the forward pass to update per-view loss statistics, and selectively skips backward passes when the sampled view's loss is consistent with its recent per-view baseline, while enforcing a minimum backward budget for stable optimization. On Mip-NeRF 360, compared to 3DGS, SkipGS reduces end-to-end training time by 23.1%, driven by a 42.0% reduction in post-densification time, with comparable reconstruction quality. Because it only changes when to backpropagate without modifying the renderer, representation, or loss, SkipGS is plug-and-play and compatible with other complementary efficiency strategies, enabling additive speedups. Code is available at https://github.com/ASU-ESIC-FAN-Lab/SkipGS.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。