1 big thing

The water got murkier and the scores went up

A figure summarising the study: five systems and four water regimes across the top, a row of murky underwater renderings of a submerged structure, and beneath them two Gaussian point clouds of a sunken car — one coherent and car-shaped labelled 3DGS, one scattered and diffuse labelled SeaSplat, each captioned with its PSNR and chamfer error

Figure: Álvarez-Tuñón and Graßhof · Research

Two underwater Gaussian splatting papers appeared on Monday, each declaring state of the art on an evaluation set it had assembled itself, neither citing the other. Olaya Álvarez-Tuñón and Stella Graßhof have now done the thing that was missing: taken five underwater splatting systems with public code, run them under a single protocol with shared poses, initialisation, compute budget and evaluator, and pushed them across four water regimes. The conclusion is that the architecture matters less than the setup, and that the numbers the field reports do not show it.

Why it matters: Underwater splatting has a benchmark culture problem that is easy to state and hard to see: the standard photometric metrics reward a model for fitting the water rather than the scene behind it. Haze is smooth, low-frequency and highly predictable, so a system that commits opacity to the veil scores well on held-out views while placing its geometry in the wrong place.

Because most published comparisons vary the architecture and the dataset together, that failure never surfaces. Running everything under one protocol separates them, and what falls out is that several confident published rankings invert depending on which water you are in and where the light is.

The Tanks and Temples Truck scene — a pale blue vintage flatbed pickup parked on a pavement — rendered sharply inside the Splat.js browser interface, with a readout showing 579,748 splats and a Train button in the toolbar

Arrival.Space gave away the browser version of what it sells

6 hours ago

Three columns comparing a reconstruction of a hand moving across a carpet — ground truth, a monocular reconstruction in which the hand dissolves into a vertical smear, and an iPhone multi-camera reconstruction in which the closed fist is legible — each shown as a wide view above a zoomed crop

Your phone records three viewpoints per shot. The pipeline throws two away.

1 hour ago

A soft, hazy Gaussian splat render of San Francisco seen from Telegraph Hill, with Coit Tower rising in the centre of the frame and the downtown skyline behind it

The framework under deck.gl just shipped a Gaussian splat renderer

10 hours ago

Two rows of underwater reef photographs, each showing a raw teal-cast frame, a SeaSplat restoration and NemoSplat's restoration, in which the water's colour cast lifts and pink and orange coral becomes visible

Where the light fails, optical flow still beats the foundation models

13 hours ago

Priya Raghunathan

17 hours agoResearch

These dynamic splats run backwards, and the round trip loses 1.3%

A grid comparing a reconstructed electric fan at t = -0.5, 0, 0.5, 1.0 and 3.0. The upper row, from conventional dynamic splatting, is marked unavailable at negative time and shows a smeared, collapsing fan at t = 3.0. The lower row, from LagrangeGS, renders the fan cleanly at every timestep with orange arcs marking particle velocity, and adds a physics-editing column

Shogo Sato and colleagues at NTT's Human Informatics Laboratories have published LagrangeGS, which takes the physics-aware dynamic splatting methods of the past two years and asks what happens if the motion is made to satisfy mechanics rather than merely resemble it. The answer is that you gain the ability to integrate the scene backwards in time, you stop objects inflating during long rollouts, and on the best available backbone your rendering metrics get worse.

Why it matters: Dynamic 3D Gaussian splatting reconstructs a moving scene beautifully inside the window it was trained on. Step outside that window and it falls apart, which is why a run of recent work has bolted physics onto it — usually by predicting a velocity field for the Gaussians and integrating forward.

Go deeper (4 min. read) ⟶

Wen Jiang

2 days agoResearch

Their filter gained 12 dB and the pictures look the same

Two rows comparing quarter-resolution renders: ground truth, baseline and filtered renders that look near-identical, followed by squared-error maps in which the baseline's error traces a bright outline around each figure and the filtered version is almost entirely black

Ankit Dhiman and colleagues at the Indian Institute of Science and Samsung R&D Bangalore have built an anti-aliasing filter for 4D Gaussian representations that adapts its strength to how fast each part of the scene is moving — and then spent a page of appendix establishing that its headline decibel numbers come almost entirely from a thin ring of pixels around object silhouettes.

Why it matters: A 4D splat is trained at one resolution and then, in any real application, viewed at others. Zoom in and each Gaussian is sampled more finely than it was fitted for; zoom out and it is sampled more coarsely. Both directions alias.

Go deeper (4 min. read) ⟶

Priya Raghunathan

2 days agoResearch

A free-flyer rebuilt the ISS interior from photos it took five years ago

Three grayscale novel views of the International Space Station's Kibō module — a stowage corridor, an airlock hatch and a crew workstation — above their corresponding rainbow-coloured depth maps

Hudson Kim of UC Berkeley, Ryan Soussan and Brian Coltin of the Intelligent Robotics Group at NASA Ames, and Jordan Kam of Caltech have published what they believe is the first 3D Gaussian splatting reconstruction of the International Space Station — built entirely from grayscale navigation images an Astrobee free-flying robot already collected in April 2021.

Why it matters: The inside of the ISS changes constantly. Cargo arrives and is stowed, payloads are swapped, cables get rerouted, equipment migrates. The reference model ground controllers work from is a static CAD file, which is accurate about what the station was designed to be and silent about where anything currently is.

Go deeper (4 min. read) ⟶

Theo Lindqvist

2 days agoTools

Three Blender splat add-ons landed in two days, each one routing around COLMAP

The PCD Splatter product image: angled panels showing an industrial site progressing from a sparse monochrome point cloud through densified grey splats to a fully coloured splat reconstruction of pipework and gantries

Between Monday and Tuesday, three separate developers shipped Gaussian splatting add-ons for Blender. None of them competes with the others, and all three are aimed at the same part of the pipeline: the structure-from-motion solve that has sat at the front of every splat workflow since 2023.

Why it matters: The standard recipe for a splat is photographs in, COLMAP solve, trainer, splat out. The solve is the slowest stage, the most likely to fail outright, and the one that assumes your input is a set of photographs of a real place taken by a camera whose position nobody wrote down.

Go deeper (3 min. read) ⟶

Priya Raghunathan

6 days agoResearch

The reward paid out highest when the video stopped moving

A grid of five video frames across ten seconds for four models, showing a cat carrying a fish along a beach; the two reward-trained baselines barely change between frames while the bottom row keeps running

Yuanhao Ban and colleagues at UCLA and Tsinghua University have identified a reward-hacking failure in streaming autoregressive video models with an unusually clean signature: the reconstruction-based critics used to keep long rollouts geometrically consistent are maximised by a video that holds still, and the models found that out.

Why it matters: Long-horizon video generation drifts. The standard fix has been to score rollouts against a 3D gaussian splatting reconstruction, on the theory that a clip which reconstructs cleanly is a clip with coherent geometry.

Go deeper (3 min. read) ⟶

Wen Jiang

6 days agoResearch

One handheld video is now enough to rebuild a person in 4D

A bass player stands at the centre of a ring of generated camera views, with a phone at lower left labelled casually captured monocular video and the surrounding frames labelled generated multiview videos

Yudong Jin, Tao Xie and colleagues at Zhejiang University's State Key Lab of CAD&CG, with Robbyant, Ant Group and HKUST, have built 4DAnyone, which takes a casually captured monocular video with unknown camera intrinsics and poses, generates the tens of multiview-consistent videos a 4D gaussian splatting reconstruction needs, and lifts them into a model you can orbit.

Why it matters: Free-viewpoint video of people has been a rig problem. DNA-Rendering, the benchmark this paper reports on, was captured with 48 synchronised cameras. That is the standard setup, and it is the reason volumetric human capture stayed inside studios that could afford one.

Go deeper (3 min. read) ⟶

View all stories ⟶