zju3dv/manhattan_sdf
Code for "Neural 3D Scene Reconstruction with the Manhattan-world Assumption" CVPR 2022 Oral
Manhattan‑SDF – 帶有曼哈頓世界先驗的神經3D重建
是什麼 – 一種從RGB‑D影像重建室內3D場景的研究程式碼庫，透過學習符號距離函數（SDF）表示來實現。它擴展了近期的神經渲染方法（NeRF、VolSDF、NeuS等），引入了曼哈頓世界假設，即大多數表面與三個正交軸對齊，這有助於正則化幾何結構，並在雜亂的室內掃描中提升重建品質。
主要元件
- 基於VolSDF框架建構的神經SDF模型，修改為強制執行軸對齊平面先驗。
- 接收多視角ScanNet資料（或自訂資料集）並優化SDF與相機位姿的訓練流程。
- 將學習到的SDF轉換為水密
.obj網格的網格提取腳本。
- 用於與COLMAP、ACMP、NeRF、UNISURF、NeuS和VolSDF等基線進行定量比較的評估工具。
快速上手
- 環境 – 執行
conda env create -f environment.yml 並透過 conda activate manhattan 激活環境。
- 資料 – 從提供的Google雲端硬碟連結下載預處理的ScanNet場景，並依照設定檔要求放置於
data/ 目錄下。
- 訓練 – 範例指令：
python train_net.py --cfg_file configs/scannet/0050.yaml gpus 0, exp_name scannet_0050
- 網格提取 – 訓練完成後，執行：
python run.py --type mesh_extract --output_mesh result.obj \
--cfg_file configs/scannet/0050.yaml gpus 0, exp_name scannet_0050
- 評估 – 使用相同的
run.py 並搭配 --type evaluate 以重現論文中的數值。
自訂資料 – 倉儲包含 docs/CUSTOM.md 指南，說明如何將此流程適配至您自己的影像集（相機內參、位姿格式等）。
為何重要 – 透過利用曼哈頓世界先驗，該方法在具有大量軸對齊牆壁與家具的場景中，相比原始神經SDF/NeRF方法，實現了更清晰、更精確的室內重建。這使其在室內地圖建構、AR/VR內容創作與機器人感知等應用中極具價值。
引用 – 若您使用此程式碼，請引用CVPR‑2022論文：
@inproceedings{guo2022manhattan,
title={Neural 3D Scene Reconstruction with the Manhattan-world Assumption},
author={Guo, Haoyu and Peng, Sida and Lin, Haotong and Wang, Qianqian and Zhang, Guofeng and Bao, Hujun and Zhou, Xiaowei},
booktitle={CVPR},
year={2022}
}
致謝 – 基於VolSDF（Lior Yariv）、COLMAP以及README中引用的多個開源實作。
相關
- 專案
- 專案
hku-mars/GS-SDF一個結合了 Gaussian Splatting 與神經符號距離場 (SDF) 的 LiDAR 增強型系統，旨在實現幾何學上一致的、照片級寫實的渲染與 ................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................thought: 3D surface reconstruction.
- 專案
- 專案