Core Concepts
Proposing a real-world dataset and benchmark for evaluating pipelines on 3D reconstruction accuracy and novel view synthesis quality.
Abstract
The MuSHRoom dataset includes 10 rooms captured by Kinect and iPhone, providing ground-truth mesh models.
Challenges include sparseness, occlusion, motion blur, reflection, transparency, and illumination variations.
Comparison methods include Nerfacto, Depth-Nerfacto, MonoSDF, and Splatfacto.
Metrics used for evaluation include accuracy, completion, Chamfer distance, normal consistency, F-score, PSNR, SSIM, and LPIPS.
Mesh culling protocol is applied to align predicted meshes with ground truth meshes.
Stats
The MuSHRoom dataset provides camera poses and point clouds for Kinect and iPhone sequences.
Quotes
"Our dataset presents exciting challenges and requires state-of-the-art methods to be cost-effective." - Xuqian Ren et al.