iiinova — where your digital self keeps evolving

This gives us a practical measure of uncertainty that we use to 1. guide captures for better data, 2. share data between points, and 3. guide data-based priors in a diffusion model. 5/6

A grid of images showing the input, the result of optimization, the result of blended diffusion and full diffusion for three objects: a plant pot, a toy car and a soda can.

100 1

Entropy can be expensive to compute. In our paper, we use frequency analysis to simplify and accelerate this computation. Our method takes 0.0005s, where it takes 6 minutes with a physically-based path tracer (Mitsuba), while still correlating strongly (0.9) 4/6

The image shows a grid of images with the input to our method, the output of entropy computed with Mitsuba and the output of entropy computed with our method. The two outputs match. The results are shown for three objects: a can, building blocks and a toy car.

100 1

How do we know when this is the case? In our paper, we use **entropy** to measure uncertainty. We argue (and show in our experiments) that where there is high entropy, we are more likely to have high error. 3/6

The figure shows a grid of images which show the input to our method, an error map and an uncertainty map. The final column contains a mask of high error and high uncertainty regions, which visually match.

First, an example of uncertainty. In the image on the left, it is hard to tell a matte plane from a glossy plane, while it's clear on the right. In the first case, better algorithms will not give a better estimate, because the data is limited. 2/6

The figure highlights material ambiguity caused by illumination. On the left, two identical gray planar surfaces are shown under an overcast environment, appearing almost indistinguishable. A spherical environment map of the overcast scene is included. On the right, the same surfaces are placed in a different outdoor environment (an asphalt road under a bright sky). Here, one surface appears significantly more mirror-like due to lighting effects. A corresponding spherical environment map of the road scene is also shown. The figure illustrates how the same material can look very different under varying lighting, complicating material estimation.

100 1

Ruben Wiersma @rubenwiersma.nl · 10/04/2025

Recent progress in multi-view 3D object capture is stunning! But is there a limit? In our upcoming #SIGGRAPH2025 paper "Uncertainty for SVBRDF Acquisition using Frequency Analysis," we show how to answer this question for material estimation. 1/6

The figure illustrates a pipeline for estimating uncertainty and spatially-varying material properties of a plush dinosaur toy. On the left, a rendered model shows the input: incoming lighting, the BRDF, and outgoing lighting from multi-view captures. Three plots in the center display spherical harmonics power spectra for incoming lighting, BRDF, and outgoing lighting. Right side: first, a colored uncertainty map (entropy) over the dinosaur surface is shown, computed within 0.5 milliseconds. Then, three images show estimated base color, roughness, and metallicity maps, computed in about 3 seconds.

1236 23

Checking the "Submission Results" email at 3AM

040 4

Ruben Wiersma @rubenwiersma.nl · 28/01/2025

Found some exploratory experiments I did early in my PhD: @sniklaus.com's 3D Ken Burns effect on paintings by Pieter de Hooch. Didn't go anywhere, but still looks cool. More below.

181 8

Ruben Wiersma @rubenwiersma.nl · 23/01/2025

Debugging results right until the deadline.

040 4

Ruben Wiersma @rubenwiersma.nl · 21/01/2025

You don't want edges on shadows in your beautiful #Blender #SIGGRAPH paper figures. An easy way to get rid of them is to go to the compositing tab, check the box 'Use Nodes' and add a 'Map Range' node on the alpha channel. Then set 'From Min' to a low value👌

0192 19

A hot glue gun was enough to link the tori into the final standing statue. If you want to read the dissertation, you can find it here :) repository.tudelft.nl/record/uuid:... 7/7

051 5

Finally, I linked the rings together by making a small cut and gluing back the missing piece. More laser cutter video, because it's so cool 😎 6/7

Etching is magical. There could have been a problem with the focal point of the laser and the curved surface, but it turned out okay in practice. 5/7

The fun part was etching text on a curved surface. I used an orthographic camera to render the scene, which models how the laser cutter 'views' the object. 4/7

100 1

The first step was to get a digital copy of the wooden rings, which was as simple as measuring the radii and creating a torus in Blender. 3/7

The cover is a set of three linked tori with the title mapped on the surface. I thought about 3D printing, but figured it would be challenging to get the overhanging parts right. I eventually went with prefab wooden rings that I would etch with a laser cutter. 2/7