COLMAP Data Gaussian Reconstruction Workflow

1.Select Project for Reconstruction

Click the "COLMAP Reconstruction" button to start the reconstruction process.

2.Parameter Settings

After closing the project, a parameter settings window will pop up. Users can configure reconstruction parameters on this interface to optimize Gaussian reconstruction quality and speed. The interface is shown below:

COLMAP重建

Parameter Settings

• GPU: Select the GPU to be used for Gaussian reconstruction. Currently, only single-GPU computation mode is supported.
• CPU: Displays the current CPU model and memory capacity of the system, allowing users to understand the system resource configuration.
• Use Sky Box: Outdoor data please select yes，Indoor data select no。
• Remaining GPU Memory: Specifies the remaining GPU memory required for the next reconstruction.
• Tiling: For data with real-world scale and large coverage areas (e.g., RTK-equipped drone captures), select 'Yes'. For data with smaller coverage (e.g., digital camera or smartphone captures), select 'No'.
• Camera Pose: Metashape和Reality Capture等软件处理后的COLMAP数据
• image Set: Image data corresponding to camera poses
• Mask（Optional）: Image data corresponding to camera poses
• Output path: Image data corresponding to camera poses

4.启动重建

​ 1)Click the "Next" → "Start" button to initiate the reconstruction function. The system will check the remaining available RAM and VRAM before launching reconstruction. If the available system memory is below 20GB, or if the selected GPU's available VRAM is lower than the minimum requirement specified in the parameter settings, Gaussian reconstruction cannot be executed.

​ 2)If the system resources meet the reconstruction requirements, the system will start the reconstruction process. The log window and progress bar will display the current processing steps and overall progress in real time.

​ 3)Click the "Open Log Folder" button to view the log records of all Gaussian reconstruction processes in the current project, allowing you to track and analyze the detailed information of each reconstruction.

​ 4)Clicking the "Stop" button will interrupt the current reconstruction process.

​ 5)After the reconstruction is successful, the system will display a prompt window asking if you want to view the reconstruction results immediately. Clicking "Yes" will open the generated reconstruction model, allowing you to browse and validate the model.

Note: During the reconstruction process, the generated point cloud slice files and log files are automatically saved to the current project's Temp and Log directories. For example, 0_1.ply is saved as the point cloud slice file, and Reconstruction_2024-10-21-13-50-03_LiDAR360MLS.log is saved as the log file for the inference process.