How we compared Teleport to other 3D scanning solutions

The challenge of capturing spaces in 3D

Teleport’s goal is to let you use any camera to capture a 3D representation of a place, in photorealistic detail. While many apps and software packages exist for scanning objects, capturing full environments is a much harder problem. In part, this is because there it requires processing a much larger amount of data. But it is also because real world spaces have complex geometries, occlusions, and un-controlled lighting conditions.

We have not yet fully solved this challenge, but we believe that we have come much closer than any other solution on the market today. In this blog, we describe how we’ve proved that Teleport is the best solution available for spatial capture using normal cameras. We realise that making head-to-head comparisons with other solutions can be a touchy business, so we want to make our approach and methodology as open, and repeatable, as possible.

Solutions compared

We compared Teleport against the following major solutions that offer Gaussian Splat reconstructions:

• Cloud-based services: Luma Labs, Polycam & Kiri Engine
• Desktop software: Jawset Postshot

These are generalist solutions, which allow any camera to scan both objects and places - but for the sake of direct comparison with Teleport we will only consider their abilities in the reconstruction of spaces.

We did not make comparisons to services with specialised cameras (Lixel, Matterport) nor to Apps with only on-device processing (Scaniverse), since we regard these as rather distinct approaches, that operate within different constraints.

Scenes used

The first step was to pick some representative datasets that we can use as standardised comparisons. From a breadth of cases, we chose:

• Protolab: This is a single room, containing many detailed objects. The data set that was professionally captured by a photogrammetry expert using a DSLR camera. It contains 1086 images, at 2500 x 1667 resolution [Download dataset]
• Atrium: This is a large, relatively featureless space. It was captured using an iPhone camera with the 0.5x lens. It contains 656 images at 1440 x 1920 resolution [Download dataset]
• Almeda House: This is a multi-room house, captured with a DSLR. It was captured as one of the datasets for the ZipNerf paper (2023). It contains 1734 images at 1394 x 793 resolution [Download dataset]

Testing method

The next step was to prepare these datasets for each of the different tools. Some of the platforms had limitations on the size of the data that could be uploaded.

• Teleport: Uploaded the un-modified images
• Luma: Uploaded the un-modified images
• Polycam: Polycam Pro plan for allows up to up 1000 images or videos of up to 15 minutes for Gaussian Splatting. We first tried combining the all images into a 15 minute video, but better results were achieved by pre-sampling every n-th image to build a 1000-image dataset. We used this method for our comparisons of Almeda & Protolab. For Atrium (656 frames), all the images were submitted.
• Kiri Engine: Kiri Engine Pro plan allows up to 300 images, or videos of up to 3 minutes for Gaussian Splatting. We found that the best results were achieved by combining all the images into a 3 minute long video, and allowing Kiri Engine to select the frames.
• Postshot: We obtained the best results by first aligning the un-modified images in Reality Capture (1.5.1), and then exporting the poses, point clouds and images into Postshot (0.6). Apart from selecting "use all images", we used the default settings.

The table below shows the links to resultant 3D reconstructions:

The resulting .plys can be downloaded here.

• Protolab
• Atrium
• Almeda

Quality assessment

To really assess the relative quality of the 3D scene, it is important to compare the scenes from exactly the same viewing angle. We did this by exporting each of the .ply files into Supersplat, and manually aligning them. By super-imposing the scenes, it is possible to make consistent view-points. We used these to create the comparative videos and sliders, such as the ones you see here.

We evaluated each reconstruction based on:

• Structural accuracy: Are floors, walls, and ceilings correctly formed?
• Visual clarity: Are surfaces sharp, are details preserved, and look photorealistic?
• Noise and artifacts: Are there floaters, haze, or major distortions?
• Handling of light and reflections: Does it preserve how the space feels?

Making side-by-side comparisons, we observed:

Protolab: A single room with dense detail

Protolab is a compact, object-rich room filled with multiple objects and varying textures, which is a good test of how well each tool handles high amount of visual detail in an indoor space.

Teleport produced a highly accurate reconstruction with minimal noise. Walls and floors were clean, surfaces aligned well, and small details like cables and signage coming through clearly. There were almost no floaters or haze. Postshot delivered similar quality to Teleport in this scene.

In contrast, Polycam, Luma and Kiri struggled with clarity. Both produced blurry textures and introduced noticeable floating artifacts. Surfaces lacked sharpness and structure.

Atrium: Large featureless space with reflections

The Atrium scene challenged each solution's ability to handle large, mostly empty spaces with plain walls, flat geometry and reflective surfaces.

Teleport produced a clean reconstruction of the walls and floors, with minimal floaters or haze. Kiri, Luma and Postshot also produced creditable reconstructions, but the Teleport scan is set apart by the flatness of the surfaces and the clarity of the reflections. Polycam had major deformations and large floaters

Almeda House: Multi-room, realistic residential layout

This was the toughest test: multiple rooms, stairs, tight corners, and uneven lighting. Teleport was the only solution to successfully reconstruct the full layout. Room transitions and overall geometry came through cleanly, with fewer floaters and artifacts.

Postshot did produce a reasonable representation, but there were significant floaters and haze. None of the other solutions, Polycam, Luma or Kiri were able to produce a well-structured reconstruction of this multi-room space.

Final thoughts: Where Teleport leads

App-based tools like Luma, Polycam, and Kiri can be effective for object captures, and for simple single-room captures. They’re quick to use and often good enough for:

• Scanning small spaces for design references
• Capturing objects or furniture for AR or 3D workflows
• Creating quick visualisations for sharing or concepting

But these tools tend to break down when the scene gets more complex in scale and detail. Large spaces, low-texture surfaces, tricky lighting, and multi-room layouts expose their limitations in geometry, alignment, artifacts, and visual fidelity.

That’s where Teleport stands out. It matches and sometimes exceeds desktop tools like Postshot, but with no manual setup, no alignment steps, and no tuning of parameters.

We're not claiming perfection. There's more to build. But right now, Teleport is the most compelling solution to capture spaces in 3D, and it's only getting better.

You can explore the interactive results at teleport.varjo.com/compare