Raindrop GS: A Benchmark for 3D Gaussian Splatting under Raindrop Conditions

ICLR 2026 Conference Submission25124 Authors

Keywords: 3D Computer Vision

Abstract: 3D Gaussian Splatting (3DGS) under raindrop conditions suffers from severe occlusions and optical distortions caused by raindrops on the camera lens, substantially degrading reconstruction quality. Existing benchmarks typically evaluate 3DGS using synthetic raindrop images with known camera poses (constrained images), assuming ideal conditions. However, in real-world scenarios, raindrops often interfere with accurate camera pose estimation and point cloud initialization. Moreover, a significant domain gap between synthetic and real raindrops further impairs generalization. To tackle these issues, we introduce RaindropGS, a comprehensive benchmark designed to evaluate the full 3DGS pipeline—from unconstrained, raindrop-corrupted images to clean 3D Gaussian reconstructions. Unlike previous benchmarks that focus solely on 3DGS reconstruction, RaindropGS enables holistic, end-to-end assessment under realistic conditions. We first collect a real-world raindrop reconstruction dataset, in which each scene contains three aligned image sets: raindrop-focused, background-focused, and rain-free ground truth, enabling comprehensive evaluation of reconstruction quality under different focus conditions. Using this dataset, we construct a complete evaluation pipeline encompassing camera pose estimation, point cloud initialization, raindrop removal preprocessing, and final 3DGS reconstruction. This modular setup allows for fine-grained analysis of each stage's impact on reconstruction quality. Through comprehensive experiments and analyses, we reveal critical insights into the performance limitations of existing 3DGS methods on unconstrained raindrop images and the varying impact of different pipeline components. These insights establish clear directions for developing more robust 3DGS methods in raindrop conditions.

Primary Area: datasets and benchmarks

Submission Number: 25124