3D Gaussian Splatting: From Photos to Explorable Spaces Visual-story transcript Runtime: 1:12 Voice: Samantha, US English, 200 words per minute 1. A captured place A photograph preserves a view. But weeks later, a reviewer needs the place around it: the doorway, the wall, and what sat behind the camera. 2. The idea 3D Gaussian Splatting offers a different record. From overlapping photographs, it estimates camera positions and builds an explorable scene from soft, coloured three-dimensional shapes. 3. Our first capture We tested the idea with an iPhone in our office. The capture looked ordinary: move through the room, keep enough overlap, and let the system track the route. 4. Where the evidence broke Then the evidence broke. Of ninety-nine sampled frames, only sixty-five registered. The first thirty-four views were missing, leaving a gap no amount of training could restore. 5. The result The result was still recognisable: about two hundred and twenty-four thousand Gaussians, rendered as a place we could move through. But recognisable did not mean complete. 6. The boundary That distinction matters in construction. A splat can help someone revisit a location and understand visual context. It is not automatically a survey, a mesh, or a Building Information Model. 7. What comes next Our next test begins with better capture: slower movement, consistent overlap, visible tracking loss, and reference dimensions. The goal is evidence whose gaps are clear. 8. Continue the inquiry Explore the full experiment in the BIMSAARTHI Community, and tell us: which site decision would benefit from better spatial context? Production note The screen uses short reinforcement phrases. The selectable caption track divides this narration into brief movie-style cues.