InfiniSplat v2: Analyzing and Improving Feed-Forward 3DGS from Any Views
InfiniSplat v2 improves feed-forward 3D Gaussian rendering quality across large viewpoint changes.
InfiniSplat v2 improves feed-forward 3D Gaussian reconstruction for novel-view rendering under large viewpoint changes. It freezes the geometry branch, aligns predicted Gaussians with ground-truth camera poses during training, and uses large-angle target-view supervision. Across six datasets, the same model reports state-of-the-art mean LPIPS and DISTS in pose-conditioned and pose-free settings. In the pose-conditioned setting, mean LPIPS and DISTS decrease by 25.8% and 32.9% versus the strongest evaluated baselines.
- InfiniSplat v2 separates frozen geometry prediction from Gaussian attribute learning.
- Training aligns Gaussians with ground-truth poses and large-angle target views.
- Mean LPIPS and DISTS fall 25.8% and 32.9% in pose-conditioned tests.
- The same model leads in both pose-conditioned and pose-free settings.
- AI models
- InfiniSplat v2
From the original article
1 min read at huggingface.co
Feed-forward Gaussian reconstruction builds 3D scenes directly from input images for novel view rendering, but learning Gaussian attributes from predicted geometry for high-quality rendering under large viewpoint changes remains challenging. We analyze limitations in Gaussian attribute learning and introduce InfiniSplat v2, which improves rendering quality through architectural and training changes.
By Yongzhen Hu, Hao Yu, Jiawei Wang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.