Gaussian splatting for adaptive reuse: define the capture acceptance test

1. Review record

An immersive record of an existing building can help a dispersed design team discuss surfaces, views and spatial sequences. Before commissioning or producing that record, define what the team must be able to learn from it and which decisions require separately measured evidence.

2. Review record

The original 3D Gaussian Splatting research, published in 2023, represents a photographed scene with optimized three-dimensional Gaussians for efficient novel-view rendering. The authors' project page links the paper, code and test scenes. It is a scene-representation and rendering method; the cited work does not establish a project's survey tolerance or certify architectural dimensions. Original research project · Paper record, submitted 8 August 2023

3. Begin with the decision the capture supports

GAD recommends a written capture brief for an adaptive-reuse pilot. Identify the rooms, elevations and routes that need discussion. Distinguish visual orientation, finish review and condition observations from setting-out dimensions or detailed construction assessment. Assign the measured information to an appropriate survey or specialist investigation.

4. Review record

Choose a small, representative area first. Include an ordinary room and a difficult junction, such as a stair meeting a landing. Define a few questions that someone unfamiliar with the building should answer from the deliverable. If the purpose is understanding circulation, check whether the transitions between spaces remain legible. If the purpose is discussing finishes, check the surfaces actually needed for that conversation.

5. Plan coverage and record its limits

Use a capture plan that marks inaccessible areas, occlusions and changing conditions. Keep a dated source-photo set and an index relating it to the building. Obtain the necessary site and image permissions, and avoid including people or confidential information unnecessarily.

6. Review record

Treat mirrors, glazing, thin elements, repetitive surfaces and moving objects as inspection targets in the resulting view. These are sensible stress cases for the proposed acceptance test, rather than a claim that every implementation fails on them. Walk the digital record along the routes the team expects to use and look for disappearance, discontinuity or ambiguous detail.

7. Test visual and measured claims separately

Have a reviewer compare selected digital viewpoints with source photographs and observations from site. Record which areas are reliable for the agreed visual purpose and which require another visit. Keep uncertainty attached to the location so that a viewer cannot mistake a gap for a confirmed condition.

8. Review record

Where a downstream workflow extracts geometry or dimensions, require its own validation against measured control and the project's agreed tolerances. Record the extraction method and its version. A visually coherent walkthrough does not by itself provide those checks, and a later conversion should not inherit approval automatically.

9. Hand over an honest record

Deliver the capture date, coverage plan, source references, viewing instructions and a short limitations note with the visual asset. Keep the measured survey and any condition assessment clearly identified alongside it. The acceptance question is whether the record supports the intended design conversation without encouraging unsupported precision. For reuse work, knowing exactly what remains unseen can be as useful as another convincing viewpoint.

Worked review scenario

Capture a representative room and stair junction; separately validate any dimensions extracted downstream.