License	Documentation	Continuous Integration	Build	Code Quality	Chat

PoSDK is designed for Spatial Visual Computing applications including camera pose estimation, feature point cloud generation, and advanced 3D reconstruction workflows. Built on innovative Pose-only Imaging Geometry theory, PoSDK provides efficient and robust solutions for lightweight, embedded vision computing applications such as robotic navigation, spatial computing, and mixed reality systems.

• Pose-only Imaging Geometry Theory: An approach that decouples pose estimation from 3D scene reconstruction, eliminating high-dimensional joint optimization
• Modular Plugin System: Flexible DataPtr/MethodPtr architecture for seamless algorithm integration and extension
• High-Performance Engine: Built-in performance profiler with real-time memory, CPU, and timing monitoring
• Automatic Evaluation System: Comprehensive accuracy evaluator with statistical analysis and CSV export capabilities
• Copyright Tracking System: Automated intellectual property protection with dependency tracking and compliance management
• Cross-platform Support: Native support for macOS (x86_64/arm64) and Linux (x86_64/aarch64)
• Platform Integration: Seamless integration and benchmarking with OpenMVG, COLMAP, GLOMAP, and other mainstream 3D reconstruction platforms

PoSDK employs a comprehensive architecture design that integrates multiple components from data preprocessing to core computation engine:

The architecture demonstrates the complete workflow from input processing through various specialized modules to final platform integration:

• Input Layer: Dynamic input processing and cross-platform integration capabilities
• Processing Modules: Image feature extraction, copyright tracking, and advanced 3D techniques (3DGS/NeRF)
• SDK Framework: Comprehensive development framework for user applications
• Core Engine: PoCore engine powered by pose-only imaging geometry algorithms
• Output Integration: Seamless integration with platforms, SDK distribution, and core services

Online Documentation

• Complete Documentation - Official documentation on ReadTheDocs
• Source Documentation - Documentation source files in GitHub

Quick Access

• User Guide - Complete introduction and feature overview
• Installation Guide - Step-by-step setup instructions
• Quick Start - Get running in minutes with precompiled binaries
• Plugin Development - Create custom algorithms and extensions
• Benchmark Results - Performance comparison with mainstream platforms
• Copyright Tracking - Automated IP protection system

Collaboration Note: This project now supports collaborative documentation editing! Contributors can directly edit Markdown files in docs/source/ and submit PRs for documentation improvements.

Local Documentation Build

# One-click setup: Install environment + Build documentation cd docs && ./install_docs.sh # View the built documentation: # English: build/html/index.html # Chinese: build_zh/html/index.html

• CMake 3.28+
• C++17 compatible compiler
• Git

# 1. Clone repository git clone https://github.com/pose-only-vision/PoSDK.git cd PoSDK # 2. One-click installation (supports resume on timeout) ./install.sh

Features:

• Resume Installation - Re-run ./install.sh if network timeouts occur
• Automated Setup - Downloads dependencies, builds PoSDK, and prepares test data For Poor Network Conditions: If automatic download fails, manually download from GitHub Releases and copy files directly (no extraction needed):
• po_core + PoSDK dependencies → ./dependencies/
• PoSDK_test_data → ./tests/

Then run ./install.sh to complete setup.

# Build PoSDK (if not done during installation) mkdir build && cd build cmake .. make -j$(nproc) # Run PoSDK from build directory ./output/bin/PoSDK

PoSDK generates comprehensive results in the working directory:

• globalsfm_pipeline_work/ - Main working directory (configured in globalsfm_pipeline.ini)
• [dataset_name]/ - Individual dataset results with 3D reconstruction and MeshLab outputs
• summary/ - Performance analysis and evaluation reports

Traditional computer vision methods often face challenges when dealing with large-scale scenes:

• Curse of dimensionality in high-dimensional optimization
• Computational inefficiency in joint pose-structure estimation
• Robustness issues in complex scenarios

Pose-only Imaging Geometry offers a paradigm shift by:

• Lossless representation of high-dimensional 3D scenes on low-dimensional pose manifolds
• Complete decoupling of camera pose estimation and 3D scene reconstruction
• Enhanced robustness without ultra-high-dimensional joint optimization

PoSDK provides comprehensive comparison with mainstream 3D reconstruction platforms:

View Detailed Benchmark Results →

PoSDK is built on peer-reviewed research in pose-only imaging geometry:

@article{cai2019equivalent, title={Equivalent Constraints for Two-View Geometry: Pose Solution/Pure Rotation Identification and 3D Reconstruction}, journal={International Journal of Computer Vision}, volume={127}, pages={163--180}, year={2019}, doi={10.1007/s11263-018-1136-9} } @article{cai2023pose, title={A Pose-only Solution to Visual Reconstruction and Navigation}, journal={IEEE Transactions on Pattern Analysis and Machine Intelligence}, volume={45}, number={1}, pages={73--86}, year={2023}, doi={10.1109/TPAMI.2021.3139681} } @article{cai2024linear, title={Linear Relative Pose Estimation Founded on Pose-only Imaging Geometry}, journal={arXiv preprint arXiv:2401.13357}, year={2024} }

If you identify any open-source license compliance issues with PoSDK, please notify us immediately through GitHub Issues. We are committed to maintaining compliance with all open-source licenses and will address any concerns promptly.

• Documentation: ReadTheDocs
• Issues: GitHub Issues
• Discussions: GitHub Discussions

Version: v1.0.0 | Architecture: Single Repository | Last Updated: 2025-01-16