DelowlightSplat: Feed-Forward Gaussian Splatting for Lowlight 3D Scene Reconstruction
For robotics and AR/VR applications requiring robust 3D reconstruction from sparse images, this work addresses the practical problem of lowlight degradation.
DelowlightSplat tackles feed-forward 3D Gaussian splatting under lowlight conditions, achieving clean novel-view synthesis despite noise and color shifts. It outperforms prior feed-forward and two-stage methods on a new lowlight benchmark.
Novel-view synthesis and 3D reconstruction from sparse posed images are central to robotics and AR/VR. Yet, feed-forward 3D Gaussian reconstruction fails under lowlight due to noise, color shifts, and unreliable correspondence. We propose DelowlightSplat, a lowlight-aware feed-forward Gaussian splatting framework for clean novel-view rendering. We build a controllable multi-view lowlight benchmark by degrading only context views while keeping target views clean. We introduce a lightweight Lowlight Adapter for residual enhancement to improve matchability, and couple it with cost-volume-based multi-view inference to directly predict clean 3D Gaussians. Experiments show that DelowlightSplat significantly outperforms previous feed-forward method and two-stage pipeline under lowlight conditions.