Pre-Publication V3. A 3D reconstruction technique turning a collection or sequence of image inputs of video inputs, such as 2D photos, into detailed, photorealistic a digital visual reconstruction 3D environments and objects scenes, representing objects and environments made up of as a high volume of millions of tiny, overlapping, semi-transparent ellipsoids ("splats"). When viewed in aggregate, these coalesce into a 3D environment or object scene. A data capture and image reconstruction technique turning a collection or sequence of image inputs into a digital visual representation (e.g. a 3D environment) made up of a high volume of overlapping, semi-transparent ellipsoids ("splats").
Deliberation Summary:
(1) Inputs are more accurately described as a collection or sequence of images, since video is not used directly but converted into frames for training. (2) Remove specific counts and size descriptors in favor of a general reference to a high volume of elements, noting that outputs may be stylized rather than strictly photorealistic, and emerging workflows extend the technique beyond static three-dimensional space to include time.
Pre-Publication V2. Gaussian Splatting (gsplats)
A 3D reconstruction technique turning a collection of video inputs, such as 2D photos, into highly detailed, photorealistic 3D scenes, . Different from traditional models built with rigid geometric shapes and polygons, it representsing objects and environments as millions of tiny, overlapping, semi-transparent ellipsoids ("splats"). When viewed in aggregate, these coalesce into a 3D scene that can very accurately represent certain things that other volumetric captures cannot: reflection, transparency, fine detail and the qualities of the light in a scene.
Deliberation Summary:
(1) Consider whether this definition distinguishes the technique from conventional geometric modeling and ensure language was not subjective. (2) Add the colloquial abbreviation to the entry title, and later forms of the technique reconstruct from multiple video inputs rather than from still images alone.
Pre-Publication V1. A 3D reconstruction technique turning a collection of video inputs, such as 2D photos, into highly detailed, photorealistic 3D scenes, . Different from traditional models built with rigid geometric shapes and polygons, it represents objects and environments as millions of tiny, overlapping, semi-transparent ellipsoids ("splats"). When viewed in aggregate, these coalesce into a 3D scene that can very accurately represent certain things that other volumetric captures cannot: reflection, transparency, fine detail and the qualities of the light in a scene.