[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-94810":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":12,"contributorsCount":14,"subscribersCount":14,"size":14,"stars1d":14,"stars7d":14,"stars30d":12,"stars90d":14,"forks30d":14,"starsTrendScore":14,"compositeScore":15,"rankGlobal":9,"rankLanguage":9,"license":16,"archived":17,"fork":17,"defaultBranch":18,"hasWiki":19,"hasPages":17,"topics":20,"createdAt":9,"pushedAt":9,"updatedAt":21,"readmeContent":22,"aiSummary":23,"trendingCount":14,"starSnapshotCount":14,"syncStatus":13,"lastSyncTime":24,"discoverSource":25},94810,"colite","maaazhuang\u002Fcolite","maaazhuang","Experimental C++17 structure-from-motion and multi-view stereo system with optional CUDA acceleration.",null,"C++",102,1,2,0,41,"BSD 3-Clause \"New\" or \"Revised\" License",false,"main",true,[],"2026-08-24 04:01:22","# Colite\n\nEnglish | [简体中文](README_zh-CN.md)\n\nColite is an experimental C++17 structure-from-motion and multi-view stereo\nsystem. It reconstructs COLMAP-compatible sparse models, RGB dense point\nclouds, Poisson or Delaunay meshes, and textured OBJ models from images.\n\nThe current release is an alpha-quality research and engineering build. It is\nprimarily developed and tested on Ubuntu 22.04 and WSL2, with optional CUDA\nacceleration for SIFT, descriptor matching, and PatchMatch Stereo. Native\nWindows builds have not yet been validated as a supported release target.\n\n## Highlights\n\n- Incremental SfM with multiple models, model continuation and robust Sim(3)\n  merging.\n- Twelve COLMAP-compatible camera models, EXIF focal initialization, optional\n  GPS position priors, and auto\u002Fsingle\u002Fper-folder\u002Fper-image intrinsics.\n- CPU and CUDA SIFT, exact CUDA brute-force descriptor matching, guided\n  matching, and several image-pair strategies.\n- CPU\u002FCUDA PatchMatch, geometric consistency, RGB\u002Fnormal Stereo Fusion,\n  Poisson and Delaunay meshing, QEM simplification, and texture atlases.\n- COLMAP text and binary sparse-model I\u002FO plus resumable stage manifests.\n- Independent model analysis, alignment, transformation, cropping, splitting,\n  bundle adjustment, triangulation, and registration commands.\n\n## Build\n\nInstall the required Ubuntu packages and build the CPU release:\n\n```bash\n.\u002Fscripts\u002Fsetup_ubuntu_22_04.sh\ncmake --preset wsl-release\ncmake --build --preset wsl-release --parallel 2\nctest --preset wsl-release\n```\n\nWith a compatible NVIDIA driver and CUDA Toolkit available:\n\n```bash\ncmake --preset wsl-colite-cuda\ncmake --build --preset wsl-colite-cuda --parallel 2\nctest --preset wsl-colite-cuda\n```\n\nThe CUDA compiler and architecture are intentionally not hard-coded. Pass\n`-DCMAKE_CUDA_COMPILER=...` or `-DCMAKE_CUDA_ARCHITECTURES=...` when automatic\ndetection is not appropriate. See [INSTALL.md](INSTALL.md) for dependencies,\ninstallation and troubleshooting.\n\n## Quick start\n\nThe repository does not distribute image datasets. Point `--image_path` at a\ndirectory containing your own images:\n\n```bash\nCOLITE=.\u002Fbuild-wsl-colite-cuda\u002Fsrc\u002Fcolite\u002Fexe\u002Fcolite\n\n$COLITE recon_auto \\\n  --image_path \u002Fpath\u002Fto\u002Fimages \\\n  --output_path output\u002Fmy_reconstruction \\\n  --sfm.max_num_features 8192 \\\n  --sfm.feature_num_threads 8 \\\n  --mvs.patch_match_backend cuda \\\n  --mesher poisson \\\n  --simplify 1 \\\n  --texture 1\n```\n\nUse the CPU build and `--mvs.patch_match_backend cpu` when CUDA is unavailable.\nInspect the exact options supported by the current build with:\n\n```bash\n$COLITE --help\n$COLITE recon_auto --help\n```\n\nThe main outputs are:\n\n```text\nOUTPUT\u002F\n  sfm\u002F                         sparse COLMAP model and RGB PLY\n  mvs\u002Fdense\u002Ffused.ply          RGB and normal dense point cloud\n  surface\u002Fmeshed-poisson.ply   or meshed-delaunay.ply\n  surface\u002Ftextured\u002Fmesh.obj    textured mesh, MTL and atlas PNG\n  .colite\u002Fstages\u002F              resume manifests\n```\n\n## Documentation\n\n- [Documentation index](docs\u002FREADME.md)\n- [Architecture and pipeline](docs\u002Farchitecture_and_pipeline.md)\n- [Command-line reference](docs\u002Fcli_reference.md)\n- [Development environment](docs\u002Fdevelopment_environment.md)\n- [Module implementation and known gaps](docs\u002Fmodule_implementation.md)\n- [Benchmark methodology and comparisons](docs\u002Fbenchmark_comparison.md)\n- [Contributing](CONTRIBUTING.md)\n\n## Current limitations\n\n- Sparse reconstruction supports incremental multi-model, global, and\n  hierarchical mapping. The global and hierarchical paths are initial\n  production implementations and still need broader large-dataset validation.\n- CUDA performance and memory usage vary substantially with image size, source\n  count, GPU architecture, and cache configuration.\n- Texture seams, visibility recall, Delaunay behavior on degenerate geometry,\n  and very large datasets need broader validation.\n- Compatibility with COLMAP formats does not imply numerical identity or\n  equivalent reconstruction accuracy on every dataset.\n\nSee the detailed known gaps in\n[docs\u002Fmodule_implementation.md](docs\u002Fmodule_implementation.md).\n\n## License\n\nColite is licensed under the [BSD 3-Clause License](LICENSE). Third-party\nnotices and retained upstream licenses are listed in\n[THIRD_PARTY_NOTICES.md](THIRD_PARTY_NOTICES.md).\n","Colite 是一个实验性的 C++17 多视图三维重建系统，支持结构光运动（SfM）与多视图立体匹配（MVS）。它可生成 COLMAP 兼容的稀疏模型、带 RGB 颜色的稠密点云、Poisson\u002FDelaunay 网格及带纹理的 OBJ 模型，并提供 CPU 与可选 CUDA 加速（用于 SIFT 特征提取、描述子匹配和 PatchMatch 立体匹配）。系统支持多种相机模型、GPS 先验、增量式 SfM、模型融合、几何一致性过滤、网格简化与纹理贴图。适用于科研验证、小规模实景三维建模及需 COLMAP 工作流兼容性的工程原型开发。","2026-08-16 02:30:08","CREATED_QUERY"]