Skip to content
  • There are no suggestions because the search field is empty.

DSM from Gaussian splatting | BETA - PIX4Dmatic

IN THIS ARTICLE

Overview
Key benefits
Known limitations
When to generate the DSM from Gaussian splatting
How to use
Orthoplanes using Gaussian splatting
Feedback

Overview

Starting with PIX4Dmatic 2.9.0 (preview), users can generate a digital surface model (DSM) directly from the Gaussian splatting model. This provides an alternative to the traditional workflow, which generates the DSM from a dense point cloud.

Note: This feature is currently in beta. Results may vary depending on your dataset and capture conditions. It is intended for experimentation and early feedback.

Key benefits

  • Sharper edges — More refined edges on buildings, vehicles, and thin structures (antennas, poles, metal frameworks).
  • Better coverage on water surfaces — The DSM based on Gaussian splatting can cover larger areas over water bodies where the dense point cloud has gaps.
  • Improved results on oblique datasets — Clear benefits for projects captured at oblique angles (e.g., tower inspections, bridge inspections, PIX4Dcatch projects).
  • Dramatically faster DSM generation — Processing is 12x to 62x faster than generating a DSM from the dense point cloud produced from Gaussian splatting. For example, a project that previously took ~35 minutes to generate a DSM may now be completed in under 2 minutes.

    Note: DTM generation is performed using terrain points and is independent of the DSM; therefore, using the DSM generated from Gaussian splatting does not affect DTM results.

    Known limitations

    • Orthomosaic quality may vary — The DSM based on Gaussian splatting can produce artifacts in the orthomosaic, particularly at building edges and on highly textured surfaces (e.g., solar panels, roof tiles). For most nadir datasets, orthomosaic quality is stable across workflows with and without Gaussian splatting.

    • Smaller spatial extent — The DSM generated from Gaussian splatting may cover a slightly smaller area than the dense-based DSM.

    • Possible holes in the DSM — Some areas may have missing data, particularly in complex scenes.

    • Visible splat artifacts — In certain cases (especially nadir or oblique PIX4Dcatch projects), artifacts from the Gaussian splats may be visible in the DSM.

    • The Gaussian splatting & point cloud step produces 2x to 4x more points and requires more processing time. While the subsequent DSM generation step is very fast, this should be considered as part of the total workflow time.

    When to generate the DSM from Gaussian splatting

    Scenario

    Recommendation

    Oblique or mixed-angle captures

    ✅ Good candidate — benefits are most visible

    Projects with thin structures (poles, antennas, vehicles)

    ✅ Good candidate — sharper edges

    Projects over water bodies

    ✅ Good candidate — better coverage

    Pure nadir datasets for orthomosaics

    ✅ Good candidate — potential artifacts at building edges and on highly textured surfaces

    How to use

    1. Process your project with Gaussian splatting enabled.
    2. In the DSM generation step, select "Gaussian splatting " as an input.
    3. Proceed with DSM and orthomosaic generation as usual.

    Orthoplanes using Gaussian splatting

    Orthoplanes can be generated from Gaussian splatting by using a section in the Section view. For the current orthoplane generation workflow, refer to the related article.

    Orthoplane - PIX4Dmatic

    Since the generation of Digital Surface Models (DSMs) from Gaussian splatting is currently in beta phase, the orthoplanes generated using this technique are also considered beta. The results may differ slightly from those obtained using the standard process.

    Feedback

    This feature is in beta, and we welcome your feedback. The more users who try it, the better we can refine the algorithm.

    Please share your experience — both positive results and issues you encounter.