WildSplat: Feedforward Gaussian Splatting from Unposed In-the-Wild Images
2026-07-06 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors introduce WildSplat, a new method for quickly creating 3D scenes from photos taken in different lighting and without known camera positions. Their approach separates the scene’s shape from its appearance to handle changing light conditions better. They use two network branches: one focused on geometry that also guesses camera positions, and one that adjusts appearance based on visual cues. Their training method helps keep these parts from mixing up. Experiments show WildSplat works better than previous methods for creating varied 3D views and editing appearances using only a few images.
3D reconstructionGaussian Splattingnovel view synthesisappearance conditioningunposed imagesdual-branch architecturecross-attentionmulti-reference trainingphotometric variationcamera pose estimation
Authors
Xiyu Zhang, Jingyu Zhuang, Hongjia Zhai, Zizheng Yan, Jinwei Chen, Guofeng Zhang, Qingnan Fan
Abstract
While feedforward 3D reconstruction excels at efficient novel view synthesis, it typically falters when faced with scenes under varying illumination. To this end, we introduce WildSplat, the first feedforward 3D Gaussian Splatting framework capable of appearance-conditioned novel-view synthesis for unposed in-the-wild images. To handle inconsistent photometric conditions, we propose a dual-branch architecture that explicitly decouples geometry from appearance. The geometry branch extracts an appearance-invariant 3D structure and jointly predicts camera poses. To govern the rendering appearance, the appearance branch injects target appearance cues into the content features via a globally pre-modulated cross-attention mechanism. To further prevent feature entanglement, we introduce a joint multi-reference training strategy that stabilizes the training process. Extensive experiments show that WildSplat surpasses existing optimization-based and feedforward methods, achieving state-of-the-art performance in in-the-wild novel view synthesis and appearance editing from sparse inputs in a single forward pass.