Reproduce “SS3DM: Benchmarking Street-View Surface Reconstruction with a Synthetic 3D Mesh Dataset” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yubin Hu, Kairui Wen, Heng Zhou, Xiaoyang Guo, Yong-Jin Liu
SS3DM is a synthetic dataset for street-view surface reconstruction, generated from the CARLA simulator, which provides precise 3D triangle mesh models as ground truth, along with multimodal sensor data. The dataset enables a robust benchmark for state-of-the-art methods, revealing challenges in scaling reconstruction to large, complex environments and highlighting the importance of evaluating surface normal quality.
Key insights
- Street-view surface reconstruction remains a highly challenging task, with performance consistently decreasing as sequence length and environmental complexity increase.
- Evaluation of surface normal quality (e.g., via Normal Chamfer Distance) is crucial for a complete assessment of reconstructed mesh fidelity and correlates better with perceptual quality than traditional distance metrics alone.
- Multi-modal inputs (RGB + LiDAR) generally lead to improved reconstruction, though optimal strategies for fusing these data streams still require further development.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.