PIX4Dmatic: controlling camera groups across multiple PIX4Dcatch datasets + 3DGS processing for SHARE3DCAM SLAM LIDAR S20 images

PIX4Dmatic
1

Dear Pix4D Support Team,

I am working on a project in PIX4Dmatic (project name: Graben_Modersohn) using data captured with PIX4Dcatch, and I would like your guidance on two separate issues.

-– Issue 1: Camera grouping across multiple PIX4Dcatch datasets —

I imported four separate PIX4Dcatch datasets into a single PIX4Dmatic project (383 images in total, all captured with iPhone 14.3 + Emlid RX, coordinate system ETRS89/DREF91/2016, EPSG:10284). The datasets cover different trench locations on the same site.

After import, PIX4Dmatic automatically split the cameras into two camera groups (“Multiple camera models” is shown at the bottom of the interface):

  • Group 1: cameras from 1 dataset
  • Group 2: cameras from the remaining 3 datasets

During calibration the four datasets are partially merged/overlapped in the reconstruction even though they correspond to physically separate excavation pits, and the resulting point cloud / Gaussian splats are fragmented (see attached screenshot).

My questions:

  1. How can I manually control the assignment of images to camera groups, so that each of the four datasets is calibrated as its own independent group?
  2. Is there a way to force PIX4Dmatic to keep the four datasets fully separated during calibration (no shared tie points, no overlapping of camera views between datasets)?
  3. What is the recommended workflow when multiple PIX4Dcatch captures of physically distinct areas should be processed together in one project without being merged?

-– Issue 2: 3D Gaussian Splatting (3DGS) for SHARE3DCAM SLAM LIDAR S20 images

I also have datasets captured with a SHARE3DCAM SLAM LIDAR S20 images. When I try to run 3D Gaussian Splatting on these images in PIX4Dmatic, the processing does not work / does not produce a 3DGS output, while the same workflow works on iPhone PIX4Dcatch data.

Could you please clarify:

  1. Is 3DGS officially supported for SHARE3DCAM SLAM LIDAR S20 images captures in the current PIX4Dmatic version?
  2. Are there specific camera model, image format, or capture settings requirements (e.g. RGB only, specific resolution, no SLAM fusion) that I need to follow for 3DGS to run on Android/S20 data?
  3. If it is supported, what logs or settings should I send for further diagnosis?

I can provide the .p4d project file, log files, and a sample of each dataset on request.

Thank you in advance for your support.

Best regards,
Humayun Sajjad
Werking Student(Geospatial Engineer)
Octagon Geotechnik GmbH

Screenshot 2026-05-26 155931
Screenshot 2026-05-26 1559311920×1036 183 KB

2

Hi @humayunsajjad.official

Let me asnwer you questions on each issue.

Issue 1.

  1. How can I manually control the assignment of images to camera groups, so that each of the four datasets is calibrated as its own independent group?

    1. The software calibrates the images in blocks during the calibration when you process multiple images in different positions. Calibrate all the images as one block is not possible since they are not connected.

  2. Is there a way to force PIX4Dmatic to keep the four datasets fully separated during calibration (no shared tie points, no overlapping of camera views between datasets)?

    1. As the dataset is not connected, the calibration process calibrate the projects in different blocks.

  3. What is the recommended workflow when multiple PIX4Dcatch captures of physically distinct areas should be processed together in one project without being merged?

    1. There is no recommended workflow to process 4 different datasets that do not contains imagen to merge them. What is the point of processing 4 different areas in the same project?

Issue 2.

  1. Is 3DGS officially supported for SHARE3DCAM SLAM LIDAR S20 images captures in the current PIX4Dmatic version?

    1. No. Fisheye cameras is not supported for 3DGS.

  2. Are there specific camera model, image format, or capture settings requirements (e.g. RGB only, specific resolution, no SLAM fusion) that I need to follow for 3DGS to run on Android/S20 data?

    1. PIX4Dmatic has an internal camera database with precalibrated models, but can process images from any camera model (RGB) as long as the images contain the necessary EXIF/XMP (metadata) tags.

  3. If it is supported, what logs or settings should I send for further diagnosis?

    1. If it is supported, the 3DGS will be generated.

I hope this information proves helpful

Regards

3

Thank you so much for your support. I go it.

4

Dear Alvaro,

Thank you for your detailed response. After reviewing your points carefully, I would like to provide further clarification on each concern to ensure we are aligned on the actual behavior we are observing.
Calibration Block Grouping
Regarding the block-based calibration, I would like to clarify that all images are indeed being processed within a single calibration run. However, the issue we are encountering is that the software is grouping all images into 2 blocks rather than the expected 4 blocks — as can be seen in the shared screenshot. This misclassification is the core concern here, and we would appreciate guidance on what may be driving this grouping behavior.

  1. Disconnected Dataset Calibration
    As referenced in the screenshot, the calibration is not splitting the project into separate blocks per disconnected area — which contradicts the behavior described in your previous response. We wanted to flag this discrepancy so it can be investigated further.

  2. Workflow for 4 Separate, Non-Overlapping Datasets
    To provide more context on our use case: this project involves a construction site where multiple trench areas have been excavated to inspect underground pipelines. Each of these areas functions as an independent mini-site. The reason we are processing all four areas within a single project is that every image carries RTK geolocation data, which assists the algorithm in accurately determining positional context. We would greatly appreciate a recommended workflow that accommodates this scenario — processing multiple spatially disconnected datasets within a single project using RTK-tagged imagery.

  3. Fisheye Camera Support for 3DGS
    We understand that fisheye cameras are currently not supported for 3D Gaussian Splatting (3DGS). However, we would like to inquire whether this is a technical limitation on the roadmap for future support, or whether there is any existing workaround — such as lens undistortion pre-processing — that could make fisheye imagery compatible with the 3DGS pipeline.

We hope these clarifications help provide a clearer picture of our workflow and the challenges we are facing. Our goal is to improve our processing pipeline, and your expertise in addressing these points would be greatly valued.

Looking forward to your response.

Best regards,
Humayun Sajjad