PoSDK Plugin List

This document lists all available plugins in the PoSDK project along with their basic information.

Plugin Function Table

Plugin Name	Input Type	Plugin Output	Function
`method_calibrator`	`data_images`	`data_cameras` (CameraModels)	Camera calibration: Uses OpenCV for pinhole camera model calibration, supports checkerboard calibration board
`method_matches_visualizer`	`data_matches`, `data_images`, `data_features`	None (visualization output to file)	Feature matching visualization: Draws two-view matching results on images, inliers shown in green, outliers in red
`method_rotation_averaging_Chatterjee`	`data_relative_poses`, `data_global_poses`	`data_global_poses` (GlobalPoses)	Global rotation averaging: Uses Chatterjee’s IRLS algorithm for global rotation estimation
`method_rotation_averaging`	`data_relative_poses`	`data_global_poses` (GlobalPoses)	Global rotation averaging: Uses GraphOptim tool for global rotation estimation
`opencv_two_view_estimator`	`data_matches`, `data_features`, `data_cameras`	`data_relative_poses` (RelativePoses)	Two-view pose estimation: Uses OpenCV to compute fundamental matrix, essential matrix, or homography matrix, supports multiple RANSAC algorithms
`opengv_model_estimator`	`data_matches`, `data_features`, `data_cameras`	`data_relative_poses` (RelativePoses)	Two-view pose estimation: Uses OpenGV library for relative pose estimation, supports RANSAC robust estimation
`openmvg_pipeline`	`data_images`	`data_images`, `data_features`, `data_matches`, `data_global_poses` (optional), `data_world_3dpoints` (optional)	OpenMVG complete pipeline: Executes complete SfM workflow, including feature extraction, matching, geometric filtering, 3D reconstruction, point cloud coloring, and quality assessment
`poselib_model_estimator`	`data_matches`, `data_features`, `data_cameras`	`data_relative_poses` (RelativePoses)	Two-view pose estimation: Uses PoseLib library for relative pose estimation
`two_view_estimator`	`data_matches`, `data_features`, `data_cameras`	`data_relative_poses` (RelativePoses)	Two-view pose estimation: Integrated PoSDK optimizer two-view pose estimation, supports multiple estimators and fine optimization
`globalsfm_pipeline`	`data_images`	`data_global_poses`, `data_world_3dpoints` (optional)	PoSDK Complete SfM Pipeline: End-to-end GlobalSfM reconstruction workflow based on PoSDK, supports multiple preprocessors and algorithm comparison

Core Pipeline Plugin Details

`globalsfm_pipeline` - PoSDK GlobalSfM Pipeline

PoSDK’s core SfM pipeline plugin, providing global 3D reconstruction solution.

Basic Configuration Parameters

Parameter Name	Type	Default Value	Description
`dataset_dir`	string	Must be set	Dataset root directory path, supports `{exe_dir}` placeholder
`image_folder`	string	`{dataset_dir}/images`	Image folder path, supports nested placeholders
`work_dir`	string	`{exe_dir}/globalsfm_pipeline_work`	Working directory path
`preprocess_type`	string	`"posdk"`	Preprocessing type: `openmvg`, `opencv`, `posdk`
`enable_evaluation`	bool	`true`	Whether to enable accuracy evaluation

Pipeline Control Options

Parameter Name	Type	Default Value	Description
`enable_3d_points_output`	bool	`true`	Whether to output 3D point cloud data
`enable_meshlab_export`	bool	`true`	Whether to export Meshlab visualization project files
`enable_posdk2colmap_export`	bool	`true`	Whether to export PoSDK reconstruction to Colmap format (cameras, images, points3D)
`enable_csv_export`	bool	`true`	Whether to enable CSV export of evaluation results
`evaluation_print_mode`	string	`"summary"`	Evaluation result print mode: `none`, `summary`, `detailed`, `comparison`
`compared_pipelines`	string	`""`	Comparison algorithm list, e.g., `"openmvg,COLMAP"`

Performance Optimization Options

Parameter Name	Type	Default Value	Description
`enable_profiling`	bool	`true`	Enable performance analysis
`enable_data_statistics`	bool	`true`	Enable data statistics functionality

Configuration File Example

Recommended configuration in {exe_dir}/../configs/methods/globalsfm_pipeline.ini:

[globalsfm_pipeline]
# Basic parameters - use placeholders for automatic adaptation
dataset_dir={exe_dir}/tests/Strecha
image_folder={dataset_dir}/castle-P19/images
work_dir={exe_dir}/globalsfm_pipeline_work

# Preprocessing and evaluation
preprocess_type=posdk
enable_evaluation=true

# Output control
enable_3d_points_output=true
enable_meshlab_export=true
enable_posdk2colmap_export=true
enable_csv_export=true
evaluation_print_mode=summary

# Performance options
enable_profiling=true

Placeholder System

Configuration supports flexible placeholder replacement:

{exe_dir}: Executable file directory
{root_dir}: Project root directory
{dataset_dir}: Dataset root directory (dynamic replacement)
{key_name}: Reference other configuration key values

Running Modes

# Default mode: use configuration file parameters
./PoSDK

# Custom mode: command-line parameters override configuration
./PoSDK --preset=custom --dataset_dir=/path/to/dataset

Output Results

Pose Data: data_global_poses - Global camera poses
3D Point Cloud: data_world_3dpoints - Sparse 3D reconstruction points (optional)
Evaluation Reports: CSV format accuracy analysis and algorithm comparison
Visualization Files: Meshlab project files (.mlp) for 3D display

For detailed parameter descriptions, please refer to comments in the configuration file.

Plugin Categories

Camera Calibration Class

method_calibrator: Pinhole camera model calibration

Visualization Class

method_matches_visualizer: Feature matching visualization

Global Rotation Estimation Class

method_rotation_averaging_Chatterjee: Chatterjee algorithm
method_rotation_averaging: GraphOptim tool

Two-View Pose Estimation Class

opencv_two_view_estimator: OpenCV implementation
opengv_model_estimator: OpenGV implementation
poselib_model_estimator: PoseLib implementation
two_view_estimator: PoSDK integrated implementation

Complete Pipeline Class

openmvg_pipeline: OpenMVG complete SfM workflow
globalsfm_pipeline: PoSDK complete SfM workflow (core recommended)

Usage Instructions

All plugins are created and used through PoSDK factory pattern:

// Create plugin instance
auto plugin = FactoryMethod::Create("plugin_name");

// Prepare input data
DataPackagePtr input_package = std::make_shared<DataPackage>();
input_package->AddData("data_type", data_ptr);

// Run plugin
DataPtr output = plugin->Build(input_package);

For detailed functionality and parameter configuration, please refer to each plugin’s source code.