This workflow demonstrates an example of performing robust and relatively fast camera alignment (SfM) using omnidirectional images: Equirectangular (equidistant cylindrical projection) images, followed by training 3D Gaussian Splatting (3DGS).
This workflow is a rewritten version of Easy&Fast 3D Gaussian Splatting workflow with 360 Camera and Metashape that replaces Metashape-based camera alignment (SfM) with COLMAP. Please read it as well.

References

Note: The examples below were created using Metashape, but similar results can be achieved with COLMAP.

DJI AVATA360 examples

• https://x.com/kotohibi\_3d/status/2079907663895482456

• https://x.com/kotohibi\_3d/status/2040724840504758578

• https://x.com/naribubu/status/2038881884558791088

• https://x.com/naribubu/status/2038875398717722743

DJI OSMO360 examples

• https://x.com/kotohibi\_3d/status/2088521899160879450

• https://x.com/kotohibi\_3d/status/2082426800215654725

• https://x.com/kotohibi\_3d/status/2074821581948481758

• https://x.com/kotohibi\_3d/status/2038179454367957106

Requirements

• 360° Camera

  • DJI OSMO360

  • DJI AVATA360

  • Insta360

• High-end PC and NVIDIA GPU

  • Training 3DGS requires a high-performance GPU. In particular, more VRAM is better. I recommend a GPU with at least 12 GB of VRAM.

• COLMAP

  • Open-source software for camera alignment (SfM).

    • https://colmap.github.io/

  • It can directly perform SfM on omnidirectional images and is relatively robust and fast.

  • This workflow uses the latest version V4.2.0 at the time of writing. You can download it from:

    • https://github.com/colmap/colmap/tags

    • I recommend the binary (CUDA) edition colmap-x64-windows-cuda.zip.

• 3D Gaussian Splatting software

  • LichtFeld Studio (LFS): https://lichtfeld.io/

  • Postshot: https://www.jawset.com/

  • Brush: https://github.com/ArthurBrussee/brush

• Still-image extraction tool from video

  • Extract Sharpest Frame (free edition)

    • https://github.com/Kotohibi/Extract\_sharpest\_frame

  • 360 Extractor (paid edition)

    • https://kotohibi.f5.si/360/extractor.html

• Cubemap conversion tool for COLMAP 360 SfM results

  • 360 CCConverter (paid edition)

    • https://kotohibi.f5.si/360/ccconverter.html

Video Shooting (e.g. OSMO360)

Attach the camera to a selfie stick and slowly walk through the area you want to capture. Recommended video settings: D-Log M, 30 fps or higher.

Develop the Video

Import the captured data into DJI Studio and perform color grading (color restoration).

• Apply the settings inside the red frame in the image below.

  

• (Advanced settings) When using the seam mask implemented in Extract Sharpest Frame V1.0.0 or later, turn RockSteady off. RockSteady provides electronic image stabilization and horizon leveling, but it changes the stitch line. Turn off the equivalent feature for Insta360 cameras as well.

  • (Note) The seam mask masks misalignments along the stitch line between the front and rear fisheye cameras, allowing that area to be excluded from the camera alignment and 3DGS training described below.

    

Export the video

• Export as an MP4 omnidirectional video. Example settings are shown in the image below.

  • Turn noise reduction on and select quality priority.

  • Turn 10-Bit Color off.

• When developing multiple clips, you can develop them together using "Multiple Clips." Extract Sharpest Frame can batch-process multiple videos.

  

Extract Still Images from Video

• There are many ways to extract still images from video. Research and choose your preferred method. Here I introduce the tool I have published. Extract Sharpest Frame is a tool that extracts the sharpest image at specified frame intervals.

• New features are prioritized for updates in the BOOTH edition

  

https://x.com/kotohibi\_3d/status/2061044432837972367
(Advanced settings) SAM3 can be configured with two types of masks: one for camera alignment and one for 3DGS training. For details, see the following Google Slides:
https://t.co/X0uRH959RV (The example uses Metashape, but the same concept applies to COLMAP.)

Execution Result

• Still images are extracted as shown below. If the SfM in the next step fails, try reducing the still-image extraction interval.

  

• Masks are also generated automatically (SAM3 + Seam mask example)

  

Perform Camera Alignment (SfM)

• Extract colmap-x64-windows-cuda.zip into any folder and run COLMAP.bat.

Create a Database and Specify the Image Files

• After running COLMAP.bat, the GUI starts.

• Select [File] -> [New Project], and the following dialog is displayed.

  

  Item	Description
  Database	Click the [New] button to create a new database file in an arbitrary location. Intermediate processing data is stored in this file.
  Images	Specify the folder containing the still images extracted from the video.
  Save	Finally, click the [Save] button to close the dialog.

Extract Features from the Images

• Select [Processing]->[Feature extraction] from the main window, and the following dialog is displayed.

  

  Item	Description
  Camera model	Select EQUIRECTANGULAR.
  Shared for all images	Turn ON.
  mask_path	Specify the created masks for camera alignment.
  use_gpu	Turn ON.
  sift.max_num_features	The default value of 8192 is also fine, but increasing it increases the initial point cloud for 3DGS training and makes training more stable. If you are using a high-spec PC, try increasing it 2–3x (or more).
  Extract	Starts feature extraction.

Run Feature Matching

• Select [Processing]->[Feature matching] from the main window, and the following dialog is displayed.

  

  Item	Description
  Type	In this guide, select the most basic SIFT_BRUTEFORCE. COLMAP allows you to choose from various feature matching algorithms. I would like to cover them in another article.
  max_num_matches	The default value of 32768 is also fine, but increasing it increases the initial point cloud for 3DGS training and makes training more stable. If you are using a high-spec PC, try increasing it 2–3x (or more).
  Run	Starts feature matching.

Run Camera Alignment

• Select [Reconstruction]->[Start reconstruction] from the main window to start camera alignment.

  

Check the Camera Alignment Result

• When processing completes successfully, you will get a result like the one below. Check that the camera positions (spherical markers) are as expected.

  

Export the Camera Alignment Result

• Select [File]->[Export model as text] from the main window to export the camera alignment information as text files.

• The following files are output to the output folder.

  

Convert to Cubemap

• Expand the COLMAP camera alignment results into 6-direction Cubemap images.
  Here I introduce the tool I have published: 360 CCConverter

Settings 1

Settings 2: Mask Processing

• You can generate masks for people, vehicles, and other objects on the expanded Cubemap images.

• If you use SAM3 masks generated by Extract Sharpest Frame, specify them in the custom mask settings below. They can be combined with YOLO masks.

  

Settings 3 (Advanced Settings): Custom Mask

• You can load SAM3 Dual Masks (for 3DGS training) generated by Extract Sharpest Frame, or any custom masks. Specify mask images as PNG files with the same count, resolution, and filenames as the still images.
  Reference: Google Slides -> https://t.co/X0uRH959RV (The example uses Metashape, but the same concept applies to COLMAP.)

  

Settings 4 (Advanced Settings): Cubemap Reduction

• This feature reduces the number of Cubemaps while minimizing 3DGS quality loss. It can shorten 3DGS training time and reduce VRAM usage.

• The ZIP file downloaded from BOOTH includes a detailed PDF operation manual. Please refer to it.

• Reference: https://x.com/kotohibi\_3d/status/2078671971639009313

  

Execute

• After processing completes successfully, the following folders and files are generated in the output folder.

  

  

(LichtFeld Studio) 3D Gaussian Splatting Training

Here I explain the workflow using LichtFeld Studio (LFS) V0.5.3.

Import Cubemap

• Select [File]-> [Import Dataset], then specify the Output Folder from CCConverter.

  

• If the data is detected correctly, a dialog like the one below appears. Confirm the contents and click [Load] to continue.

  

• Once the data is loaded correctly, you will see a screen like the one below.

  

Start 3DGS Training

• Mask settings

  • Select [Training Parameters]->[Mask Mode]->[Ignore].

  • Turn off [Alpha Mask].

• Training parameters

  • Here is an example of the settings I often use.

  • I recommend [Strategy]->[MRNF] (at the time this article was written).

  • Adjust Max Gaussians according to the scale of the scene (3,000,000–12,000,000).

  • Adjust SH Degree (1–3). If VRAM is limited, I recommend 1.

  • Iterations and Steps Scaler are calculated automatically according to the number of images.

  • With MRNF, changes to the other parameters are usually not very necessary.

  • LFS has many parameters, so please research on the web and find the best settings for your scene.

  • Note: Recently, Bilateral Grid is often turned off (PPISP is sufficient in many cases).

    

• Click [Start Training] to begin 3DGS training.

3DGS Training Result

• As training progresses, you should start seeing the 3DGS!

  

Finally

There are many 3DGS methods, and this article is just one example. I will continue sharing the latest information on my X account. Please research on your own and develop even better techniques. Enjoy 3DGS :)

• my 360 Tools: https://kotohibi.f5.si/360

• my X: https://x.com/kotohibi\_3d

• 3DGS pipeline guide: https://github.com/Kotohibi/3DGS\_pipeline\_guide

This workflow demonstrates an example of performing robust and relatively fast camera alignment (SfM) using omnidirectional images: Equirectangular (equidistant cylindrical projection) images, followed by training 3D Gaussian Splatting (3DGS).

References

DJI AVATA360 examples

• https://x.com/kotohibi\_3d/status/2079907663895482456

• https://x.com/kotohibi\_3d/status/2040724840504758578

• https://x.com/naribubu/status/2038881884558791088

• https://x.com/naribubu/status/2038875398717722743

DJI OSMO360 examples

• https://x.com/kotohibi\_3d/status/2088521899160879450

• https://x.com/kotohibi\_3d/status/2082426800215654725

• https://x.com/kotohibi\_3d/status/2074821581948481758

• https://x.com/kotohibi\_3d/status/2038179454367957106

Requirements

• 360° Camera

  • DJI OSMO360

  • DJI AVATA360

  • Insta360

• High-end PC and NVIDIA GPU

  • Training 3DGS requires a high-performance GPU. In particular, more VRAM is better. I recommend a GPU with at least 12 GB of VRAM.

• Metashape Standard (Not supported the Professional edition)

  • Directly supports SfM with omnidirectional images; extremely fast and robust.

  • https://www.agisoft.com/features/standard-edition/

• 3D Gaussian Splatting software

  • Postshot: https://www.jawset.com/

  • LichtFeld Studio (LFS): https://github.com/MrNeRF/LichtFeld-Studio

  • Brush: https://github.com/ArthurBrussee/brush

• Still-image extraction tool from video

  • Extract Sharpest Frame (Free edition)

    • https://github.com/Kotohibi/Extract\_sharpest\_frame

  • 360 Extractor (paid edition)

    • https://kotohibi.f5.si/360/

• Metashape 360 SfM to COLMAP-format Cubemap conversion tool

  • Metashape 360 to COLMAP Converter (Free edition)

    • https://github.com/Kotohibi/Metashape\_360\_to\_COLMAP\_plane

  • 360 MCConverter (paid edition)

    • https://kotohibi.f5.si/360/

• (Optional) Additional license for estimating real-world scale in 3DCG

  • Metashape 360 to COLMAP Converter includes a feature that estimates the real-world scale of 3DGS using a 2D marker called AprilTag. An additional license is required to use it. The following articles explain how to use AprilTag.

  • Add-on Real Scale 3DGS with AprilTag https://kotohibi-cg.booth.pm/items/8323677

  • English version: https://x.gd/CoWJA

  • Japanese version: https://x.gd/Isahb

Video Shooting (e.g. OSMO360)

Attach the camera to a selfie stick and slowly walk through the area you want to capture.
Recommended video settings: D-Log M, 30 fps or higher.

Develop the Video

Import the captured data into DJI Studio and perform color grading (color restoration).

• Apply the settings inside the red frame in the image below. Everything else can be left at default.

  

• (Advanced settings) When using the seam mask implemented in Extract Sharpest Frame V1.0.0 or later, turn RockSteady off. RockSteady provides electronic image stabilization and horizon leveling, but it changes the stitch line. Turn off the equivalent feature for Insta360 cameras as well.

  • (Note) The seam mask masks misalignments along the stitch line between the front and rear fisheye cameras, allowing that area to be excluded from the camera alignment and 3DGS training described below.

Export the video

• Export as an MP4 omnidirectional video. Example settings are shown in the image below.

• When developing multiple clips, you can develop them together using "Multiple Clips." Extract Sharpest Frame can batch-process multiple videos.

  

Extract Still Images from Video

There are many ways to extract still images from video. Research and choose your preferred method.
Here I introduce the tool I have published.
Extract Sharpest Frame is a tool that extracts the sharpest image at specified frame intervals.

• New features are prioritized for updates in the BOOTH edition

  

https://x.com/kotohibi\_3d/status/2061044432837972367
(Advanced settings) Two SAM3 masks can be configured separately: one for camera alignment and one for 3DGS training. For details, see the following Google Slides:
https://t.co/X0uRH959RV

Execution Result

• Still images are extracted as shown below. Please check the SfM result in the next step and readjust the Chunk size if necessary.

  

• Masks are also generated automatically

  

Perform Camera Alignment (SfM)

Use Metashape Standard, which can directly process omnidirectional images for SfM.

Load the extracted omnidirectional images

[Workflow] → [Add Folder]

Change Camera Type to Spherical

Select [Tools] → [Camera Calibration] and set Camera type to Spherical.

Load the mask images

Select [File] → [Import] → [Import Masks].

Use the settings below and click [OK]. Then a folder selection dialog will appear; select the mask folder generated by Extract Sharpest Frame.

• Set Filename template to {filename}.png (the default is {filename}_mask.png).

  

Set SfM parameters

Here are two parameter examples I often use.

Execute

Click OK to run SfM.
Example result shown below. The spherical markers correspond to each omnidirectional image.

Clean up Tie points

• Remove low-reliability Tie points to improve SfM accuracy. This is a very important step for high-detail 3DGS. Select [Tools] → [Tie Points] → [Clean Tie points].

  

• Select [Reprojection error] and adjust the slider to remove about 5% of the Tie points.

  

• The number of Tie points is shown at the bottom left of the screen, so adjust the slider while checking how many Tie points will be removed.

  

• Click [Optimize Cameras] to optimize the cameras.

  

• Do the same for [Recostruction uncertainty], remove about 5% of the Tie points, and then run [Optimize Cameras] again.

• Do the same for [Projection accuracy], remove about 5% of the Tie points, and then run [Optimize Cameras] again.

• Repeat the above once more so that low-reliability Tie points are removed as much as possible.

Export SfM results

* Export Camera information  
  [File] → [Export] → [Export Cameras] → Select Agisoft XML (*.xml) and save.
* Export Point Cloud  
  [File] → [Export] → [Export Point Cloud] → Select Stanford PLY (*.ply) and save.

Convert to COLMAP Cubemap

Convert the Metashape SfM results into COLMAP-format 6-direction Cubemap images.
Here I introduce the tool I have published.
Metashape 360 to COLMAP Converter

• New features are prioritized for updates in the BOOTH edition

Settings ①

Settings ② : Mask Processing

• If you use SAM3 masks generated by Extract Sharpest Frame, turn this feature off and specify the SAM3 masks in the custom-mask settings below.

• You can generate masks for people, vehicles, and other objects. Mask generation is particularly important with 360° cameras because the operator is often captured in the frame.

  

Settings ③ (Advanced Settings): Custom Mask

• You can load SAM3 Dual Masks (for 3DGS training) generated by Extract Sharpest Frame, or custom masks. Specify mask images as PNG files with the same file count, resolution, and filenames as the still images.
  Reference: Google Slides -> https://t.co/X0uRH959RV

  

Settings ④ (Advanced Settings): AprilTag

• This feature estimates the real-world scale of 3DGS using 2D markers. An additional license is required. The following articles explain how to use AprilTag.

  • English version: https://x.gd/CoWJA

  • Japanese version: https://x.gd/Isahb

    

Settings ⑤ (Advanced Settings): Overlap Filter

• See the [EN]Mixed 3DGS Workflow with 360 Video and Planar Images guide.

• The ZIP file downloaded from BOOTH includes a detailed PDF operation manual. Please refer to it.

• Reference: https://x.com/kotohibi\_3d/status/2078088377371681237

  

Settings ⑥ (Advanced Settings): Cubemap Reduction

• This feature reduces the number of Cubemaps while minimizing 3DGS quality loss. It can shorten 3DGS training time and reduce VRAM usage.

• The ZIP file downloaded from BOOTH includes a detailed PDF operation manual. Please refer to it.

• Reference: https://x.com/kotohibi\_3d/status/2086408535768973659

  

Execute

After processing completes successfully, the following folders and files are generated in the output folder.

(Postshot) 3D Gaussian Splatting Training

Here I explain the workflow using Postshot.

Import Cubemap

• First, drag & drop the Images folder, cameras.txt, images.txt, and points3D.txt into Postshot.

Mask Settings

• Next, drag & drop the masks folder into the Image Masks area in Postshot.
  Select Remove Background for Mask Mode.

Cubemap Import Result

• Once the Cubemap is successfully imported, you will see a screen like the one above.

Start 3DGS Training

• Here is an example of training parameters I use for wide-area 3DGS. Adjust parameters according to your scene.

• Turn on Photometric Compensation.

• Change Max Splat Count according to the scale of the scene (3,000-12,000 kSplats).

• Adjust Max Sph.Hrm.Degree (1-3). If VRAM is limited, I recommend 1.

• Stop Training After is calculated automatically according to the number of images.

  

3DGS Training Result

• As training progresses, you should start seeing the 3DGS!

(LichtFeld Studio) 3D Gaussian Splatting Training

Here I explain the workflow using LichtFeld Studio (LFS).

Import Cubemap

• Select [File] → [Import Dataset], then specify the Output Folder from Metashape 360 to COLMAP Converter.

  

• If the data is detected correctly, a dialog like the one below appears. Confirm the contents and click [Load] to continue.

  

• Once the data is loaded correctly, you will see a screen like the one below.

  

Start 3DGS Training

• Mask settings

  • Select [Training Parameters] → [Mask Mode] → [Ignore].

  • Turn off [Alpha Mask].

• Training parameters

  • Here is an example of settings I often use.

  • I recommend [Strategy] → [MRNF] (at the time this article was written).

  • Adjust Max Gaussians according to the scale of the scene (3,000,000-12,000,000).

  • Adjust SH Degree (1-3). If VRAM is limited, I recommend 1.

  • Iterations and Steps Scaler are calculated automatically according to the number of images.

  • With MRNF, changes to the other parameters are usually not very necessary.

  • LFS has many parameters, so please research on the web and find the best settings for your scene.

  • Recently, Bilateral Grid is often turned off because PPISP is sufficient in many cases.

    

• Click [Start Training] to begin 3DGS training.

3DGS Training Result

• As training progresses, you should start seeing the 3DGS!

  

Finally

There are many 3DGS methods, and this article is just one example. I will continue sharing the latest information on my X account.
Please research on your own and develop even better techniques. Enjoy 3DGS :)

• my X: https://x.com/kotohibi\_3d

• 3DGS pipeline guide: https://github.com/Kotohibi/3DGS\_pipeline\_guide

This document is intended for users who:

• Understand the following guide:

  • [EN]Easy&Fast 3D Gaussian Splatting workflow with 360 Camera

• Have a general understanding of the 3DGS production workflow using a 360 camera

• Want to take the next step and create higher-quality 3DGS results

Overview

Creating 3DGS with 360 cameras such as OSMO360 or AVATA360 is easy, but there are limits in terms of quality. On the other hand, workflows using planar images from mirrorless cameras and similar devices can achieve higher quality, but they require more shooting effort and make it difficult to reconstruct the entire scene. This workflow proposes a mixed workflow using both a 360 camera and a planar camera. The main goal is to improve 3DGS results by capturing the entire scene with a 360 camera while capturing the area of interest in detail with a planar camera.

References

Requirements

• 360 camera

  • DJI OSMO 360

  • DJI AVATA 360

  • Insta 360

  • etc...

• Planar camera

  • Mirrorless camera

  • Action camera

  • etc...

• High-end PC and NVIDIA GPU

  • Training 3DGS requires a high-performance GPU. In particular, more VRAM is better. I recommend a GPU with at least 12 GB of VRAM.

• Metashape Standard (Professional edition is not supported)

  • Directly supports SfM with 360 images; extremely fast and robust.

  • https://www.agisoft.com/features/standard-edition/

• 3D Gaussian Splatting software

  • Postshot: https://www.jawset.com/

  • LichtFeld Studio (LFS): https://github.com/MrNeRF/LichtFeld-Studio

  • Brush: https://github.com/ArthurBrussee/brush

• Still-image extraction tool from video

  • Extract Sharpest Frame (free edition) * https://github.com/Kotohibi/Extract\_sharpest\_frame

  • 360 Extractor (paid edition) * https://kotohibi.f5.si/360/

• Metashape 360 SfM to COLMAP-format Cubemap conversion tool

  • Metashape 360 to COLMAP Converter (free edition) * https://github.com/Kotohibi/Metashape\_360\_to\_COLMAP\_plane

  • 360 MCConverter (paid edition) * https://kotohibi.f5.si/360/

• (Optional) Additional license for estimating real-world scale in 3DCG

  • Metashape 360 to COLMAP Converter includes a feature that estimates the real-world scale of 3DGS using a 2D marker called AprilTag. An additional license is required to use it. The following articles explain how to use AprilTag.

  • Add-on Real Scale 3DGS with AprilTag https://kotohibi-cg.booth.pm/items/8323677

  • English version: https://x.gd/CoWJA

  • Japanese version: https://x.gd/Isahb

How This Guide Explains the Procedure

The following sections mainly explain the processing flow. For detailed tool usage and option descriptions, refer to the following guide:

• [EN]Easy&Fast 3D Gaussian Splatting workflow with 360 Camera

Preparing Videos and Still Images

This article explains 3DGS production using the following video materials as an example.
I strongly recommend matching exposure and color tone between the 360 camera and planar camera in development software beforehand. If exposure and color tone differ significantly, the 3DGS result will be worse.

Extracting Still Images and Generating Masks from 360 Video

There are many ways to extract still images from video. Research and choose your preferred method. Here I introduce the tool I have published. Extract Sharpest Frame is a tool that extracts the sharpest image at specified frame intervals. New features are prioritized for updates in the BOOTH edition

Preparing a Custom Mask (Optional)

In this example, the 360 camera is mounted under a drone, so the drone body appears in the upper half of the image. In such cases, create an image that masks the upper half and register it as a custom mask. The specified custom mask is automatically merged with YOLO and SAM3 masks. The custom mask must have the same image size as the original video. If you are using a drone with a built-in 360 camera such as DJI AVATA360, this step is not necessary.

Loading the Video

Load the 360 video in Extract Sharpest Frame. This tool can batch-process multiple videos.

• If you turn on the "collect outputs from multiple videos into one folder" option, a sequential prefix is added to each video file, and the extracted still images are stored in one folder. Masks are also stored in one folder in the same way.

  • Example still-image filenames when processing two video files

    • Video 1 | Mask: 001_[output filename pattern], ...

    • Video 2 | Mask: 002_[output filename pattern], ...

• The still-image extraction settings are shared. If you want to extract each video file with different conditions, run the extraction process multiple times.

• In that case, by using the same Output folder and changing the Output pattern, you can avoid overwriting extracted files and collect them into the same folder.

• You can limit the target timecode with "Start" and "End", which is useful for test runs.

  

SAM3 Mask Settings

This tool can configure two types of SAM3 masks. The "SAM3 Mask 1" and "SAM3 Mask 2" tabs correspond to this feature. By separating masks for camera alignment and masks for 3DGS training, you can generate higher-quality 3DGS.

Extracting Still Images and Generating Masks from Planar Video

Next, extract still images and generate masks from the planar video. The method is the same as for the 360 video. In this article, the planar video is treated as a single file from a drone-mounted camera. Because the still images and masks should be collected into the same folder as the 360 processing results, specify the same output folder. Also, to avoid overwriting files, change the [Output filename pattern] to a non-overlapping name as shown below.

Tips

Extract Sharpest Frame can create masks not only from videos, but also from still images. In that case, turn on the following [Still Image Mask Mode] and select the still-image folder.

Folder Structure

After still-image extraction and mask generation, the folder structure is as follows.

Performing Camera Alignment

Next, perform camera alignment using Metashape Standard. Metashape can align still images extracted from both 360 video and planar video at once.

Loading Still Images

• Select [Workflow] -> [Add Folder], then specify the "frames" folder above and load the extracted still images.

  

Changing Camera Type

• Select [Tools] -> [Camera Calibration].

  

If the 360 images and planar images are loaded correctly, they are automatically separated into two image groups. Set "Camera type" to "Spherical" and "Frame" respectively.

Running Camera Alignment

• Run it from [Workflow] -> [Align Photos].

SfM Parameter Settings

Here are two parameter examples I often use.

Checking the Result

If camera alignment succeeds, a result containing both 360 and planar images is displayed as shown below.

Clean up Tie points

• Remove low-reliability Tie points to improve SfM accuracy. This is a very important step for high-detail 3DGS. Select [Tools] → [Tie Points] → [Clean Tie points].

  

• Select [Reprojection error] and adjust the slider to remove about 5% of the Tie points.

  

• The number of Tie points is shown at the bottom left of the screen, so adjust the slider while checking how many Tie points will be removed.

  

• Click [Optimize Cameras] to optimize the cameras.

  

• Do the same for [Recostruction uncertainty], remove about 5% of the Tie points, and then run [Optimize Cameras] again.

• Do the same for [Projection accuracy], remove about 5% of the Tie points, and then run [Optimize Cameras] again.

• Repeat the above once more so that low-reliability Tie points are removed as much as possible.

Exporting SfM Results

• Export Camera information Select [File] -> [Export] -> [Export Cameras], choose Agisoft XML (*.xml), and save.

• Export Point Cloud Select [File] -> [Export] -> [Export Point Cloud], choose Stanford PLY (*.ply), and save.

Converting to COLMAP Cubemap

• Expand the Metashape SfM results into COLMAP-format 6-direction Cubemap images. Here I introduce the tool I have published. Metashape 360 to COLMAP Converter

• New features are prioritized for updates in the BOOTH edition

The basic usage is the same as in the guide below, but this time the 3DGS training masks generated earlier are used for Cubemap expansion.

• [EN]Easy&Fast 3D Gaussian Splatting workflow with 360 Camera

Setting a Custom Mask

From the Custom Mask tab, set the folder path of the 3DGS training masks generated during still-image extraction.

Running the Conversion

Press "Run Conversion" to start processing. At this point, a dialog appears indicating that planar images are included, but press "OK" to continue processing. 360 images are expanded into Cubemaps, while planar images are automatically undistorted using the pinhole model. Masks are also processed automatically in the same way.

Advanced

Starting with Metashape 360 to COLMAP Converter V1.2.0, the Overlap Filter (formerly Cubemap Filter) has been implemented. It detects overlap between Cubemaps expanded from 360 images and planar images, and can exclude the corresponding Cubemap faces. This increases the frequency with which the target area is trained from planar images, making it possible to improve the overall quality of 3DGS. For details, refer to the PDF included in the tool zip file. See my post on X: https://x.com/kotohibi\_3d/status/2078088377371681237

• When "Cubemap Filter" is turned ON, overlapping Cubemap regions are excluded.

  

• Result after excluding Cubemap faces.

  

Training 3DGS

Because the usage is the same as below, detailed steps are omitted.

• [EN]Easy&Fast 3D Gaussian Splatting workflow with 360 Camera

Checking the COLMAP Dataset Loading Result

3DGS Training Result

As shown below, the quality of the area of interest captured with the planar camera improved.

See my post on X: https://x.com/kotohibi\_3d/status/2048060928579850578

Discussion