GSReconstruction Module FAQ
General Questions
1.GSReconstruction Model Looking Blurry?
- Ensure that LiDAR360MLS and LiDAR360MLS_DL_Server are the latest versions downloaded from the official website.
- Ensure that the reconstructed model corresponds to the panoramic camera video and the point cloud project collected (pay attention to the path where the video is stored and confirm it is correct when selecting the video).
- If the model looks great from one angle and not from another, it could be that data was collected with the camera facing one way only, and the scanning was not performed from multiple angles, resulting in insufficient information.
- The camera lens is obstructed, for example, by RTK cables or contaminants.
- Review the recorded video to check for reflections or overexposure in the scene.
2.Does GSReconstruction Model Support OBJ and OSGB?
- GSReconstruction models are not traditional mesh surface models and are not suitable for conversion to common 3D Mesh formats such as OBJ or OSGB
3.What devices are currently supported?
- Currently, O1_Lite\O2_Lite and H300 series devices are supported.
4.How long is the data collection time in general?
- Generally, at least 5 minutes and less than 30 minutes of data should be collected. Reconstruction is not possible if the data collection time is too short or too long.
5.There are many artifacts in the reconstruction.
- The acquisition site should not have a high number of moving targets, and it is better to choose a time period when there are fewer moving targets to acquire.
6.When reconstructing using a computer with an Intel 13th or 14th generation CPU, the reconstruction fails.
- Intel 13th or 14th generation CPUs are prone to crashes caused by CPU overclocking. If re-executing the reconstruction process does not resolve the issue, please contact relevant technical personnel for support or use another computer to perform the reconstruction.
7.Before performing Gaussian reconstruction, ensure that Dynamic Object Removal, Smoothing Filter, and Auto Mask are selected during processing. If moving objects still appear in the processed point cloud, manually classify them in the processing project.
8.Does Gaussian Reconstruction support reconstruction using data in the foot coordinate system?
- Currently, Gaussian Reconstruction does not support data in the foot coordinate system. Please perform solving and reconstruction using data in the meter coordinate system.
Requirements for Running Gaussian Reconstruction
Memory (RAM): 32GB or higher.
Central Processing Unit (CPU): Intel® Core™ i5/i7 is recommended; eight-core, sixteen-thread processor with single-core performance of 3.7GHz or higher.
Graphics Card: A high-performance NVIDIA graphics card is recommended, with at least 8GB of memory. Recommended models are as follows:
Server: Tesla V100, A100, H800
- Desktop: 2080Ti, 3060, 3080Ti, 4080, 3090, 4090
Mobile: 3080Ti, 4080Ti
Gaussian Reconstruction Environment Check Errors and Solutions
1.Please update the NVIDIA graphics card driver (requires > 465.89). You can download the latest driver from (graphics card driver link).
Error Reason: The computer does not have an NVIDIA graphics card, the driver is not installed, or the installed driver version is lower than 465.89.
Solution: Use an NVIDIA graphics card or install the latest graphics driver.
2.Warning: Insufficient GPU memory. To run the reconstruction, a graphic card with a minimum of 8GB of VRAM are required.
Error Reason: The NVIDIA graphics card in the computer has less than 8GB of VRAM.
Solution: Use an NVIDIA graphics card with more than 8GB of VRAM.
3.Warning: Insufficient memory. To run the reconstruction, a system with a minimum of 32GB of RAM are required.
Error Reason: The computer has less than 32GB of RAM.
Solution: Use a computer with more than 32GB of RAM.
4.The available GPU memory is insufficient for reconstruction. Please try closing other programs to free up GPU memory.
Error Reason: The available GPU memory is insufficient.
Solution: Close other applications to free up GPU memory or select a lower memory requirement.
5.LiDAR360MLS only supports running one instance processor.
Error Reason: Another deep learning software process (e.g., gvSplat.exe, gvCVEngine.exe, gvPCEngine.exe) is running in the background.
Solution: Wait for the other deep learning software to complete or terminate the process using Task Manager.
Error Messages and Solutions When Opening a Reconstruction Project
1.[GSReconstruction]Load project file :F:/Data/0625/2024-06-25-07-12-28/2024-06-25-07-12-28.liscan failed!
Error Reason: The selected project file is corrupted.
Solution: Select a valid project file to proceed with Gaussian reconstruction.
2.[GSReconstruction]Reconstruction do not support this platform!
- Error Reason: The selected liscan file platform is not of the SLAM type.
- Solution: Select a liscan file with a SLAM-type platform for Gaussian reconstruction.
3.[GSReconstruction] Open file D:\Data\2023-08-17-16-51-37\SLAMProcess\temp\device.info failed!
Error Reason: The file storing device information for the selected liscan file is missing.
Solution: Use the Open Georeference Project button in the software to open the liscan file, re-calculate and reassign colors before proceeding with Gaussian reconstruction.
4.[GSReconstruction] Reconstruction only support O1_Lite or H300 device!
Error Reason: The selected liscan file device model does not match O1_Lite or H300.
Solution: Select a liscan file with a device model of O1_Lite or H300 for Gaussian reconstruction.
5.[GSReconstruction] Project do not have point cloud!
- Error Reason: The selected liscan file does not contain any point cloud data.
- Solution: Open the liscan file using the Open Georeference Project button in the software, recalculate and reassign colors, and then proceed with Gaussian reconstruction.
6.[GSReconstruction] Project do not have have imglist!
Error Reason: The selected liscan file does not contain any images.
Solution: Open the liscan file using the Open Georeference Project button in the software, recalculate and reassign colors, and then proceed with Gaussian reconstruction.
7.[GSReconstruction] Selected camera folder is invaild! F:/Data/0625/2024-06-25-07-12-28/Cam
Error Reason: The selected camera folder does not contain a valid original camera folder.
Solution: Select a valid original camera folder. If the liscan collection device is O1_Lite, the camera folder must contain two or more
.binfiles, and one of them must beimage_time.bin. If the liscan collection device is H300, the camera folder must contain two or more.insvfiles.
Error message and solution when running Gaussian reconstruction
1.[Downsample] Failed to load the point cloud, the point cloud file may not exist or could be missing RGB information.
Error reason:The point cloud corresponding to the liscan file does not exist or lacks RGB information.
Solution:If the point cloud corresponding to the liscan file exists, click the Open Calculation Project button to open the liscan file, reassign the colors, and then proceed with Gaussian reconstruction.
If the point cloud corresponding to the liscan file does not exist, click the Open Calculation Project button to open the liscan file, calculate and assign colors, and then proceed with Gaussian reconstruction.
2.[GSReconstruction] C:\Program Files\GreenValley Suite\LiDAR360MLS\8.1.0.0\plugins\DLModels\reconstruction.json dosen't exist!
Error reason:The Gaussian reconstruction parameter template file is missing in the Deep Learning module.
Solution:Reinstall the Deep Learning module following theInstallation of Deep Learning Extension Modules.
3.[GSReconstruction] Images path is invalid: *.jpg
- Error reason:When running Gaussian reconstruction, the image file to be parsed is missing.
- Solution: Please check if there is enough disk space and then re-run the function.
4.[GSReconstruction] offset file is invalid: *.txt
- Error reason:The offset file output by the SFM during Gaussian reconstruction is invalid.
- Solution:Please check if the reconstruction project is complete, review the logs to ensure the SFM function executed correctly, and make sure not to manually delete this file during software execution.
5.[GSReconstruction] GPU out of memory. Please use a GPU with more memory.
- Error reason:Insufficient GPU memory when running the Gaussian reconstruction function.
- Solution: When running Gaussian reconstruction, try not to use other functions on the computer, especially other memory-intensive software, as it can reduce the available GPU memory for this function.
6.[GSReconstruction] Out of memory. Too much input data or memory is too small.
- Error reason:Insufficient memory when running the Gaussian reconstruction function.
- Solution: When running Gaussian reconstruction, try not to use other functions on the computer, especially other memory-intensive software, as it can reduce the available memory for this function.
7.[GSReconstruction] LiSplat file is invalid.
- Error reason:The model file created after the Gaussian reconstruction is missing
- Solution: Ensure that there is enough disk space and that files are not manually deleted. Restart the function. Also, make sure the data collection time is longer than 5 minutes.
8.[GSReconstruction] GPU driver version in too old. Please update to version 465.89 or above.
- Error reason:The GPU driver version on the computer is too outdated.
- Solution: Download the latest GPU driver and update it.
9.[GSReconstruction] SFM result invalid.
- Error reason:Too few matching points for camera poses after SFM reconstruction.
- Solution: Follow the data collection guidelines in the manual.
10.[GSReconstruction] splitPly path is invalid:
- Error reason:The point cloud file is invalid.
- Solution: Check the point cloud files (.lidata) in the project to ensure they are valid. If not, recompute the data.
11.[GSReconstruction] depth_params.json file not found at path: *.json
- Error reason:Missing *.json file.
- Solution: Ensure that the file has not been manually deleted and rerun the program.