News from Material Appearance 2020
The conference "Material Appearance 2020", hosted by the Electronic
Imaging Symposium in Burlingame, CA, USA on the 27th and 28th January
2020, was exciting for the number of different topics addressed,
including measurement and modelization issues, physical and digital
methods, and a variety of materials and appearance attributes. A short
summary by Mathieu Hebert, chair of the conference, is delivered here.
Measurement
- 3D scanning: Scanning the 3D shape of objects at very high resolution is of high importance in the cultural heritage domain. Martin Ritz (invited speaker) presented various generations of scanners at Fraunhofer IGD, Germany. He insisted on the necessity to acquire, in addition to the 3D shape, the optical behaviour of the surface in order to achieve good rendering when the reconstructed object is diplayed on digital support. Issues related to texture mapping are also addressed. Impressive images of metallized objects were shown.
- Multispectral image devices are still developed, while trying to improve the acquisition performences and accuracy, as shown by Axel Clouet from CEA-LETI, France, who also announced the creation of a spectral databased for labelled objects.
From Measurement to Prediction
- BRDF: The microfacet model on which rely most BRDF
models for light scattering by rough surfaces is usually based on one
roughness parameter, the standard deviation of facet slope. In the case
of very glossy surfaces, this parameter is difficult to estimate from
BRDF measurements because of the apparatus function of the instrument,
unless a very high precision goniophotometer is used. Shoji Tominaga from NTNU, Norway,
explained that it is preferrable to use a glossmeter to obtain more
accurate roughness parameter estimation, and btter matching between BRDF
model and measurement..
- BTF: Naoki Tada from Chiba University, Japan,
showed how CNN can be used to predict the variation of bi-direction
texture function (BTF) according to the position of the light source.
The neural network is trained and used for various texture categories,
separate from each other.
- Renewing: Is it possible, from a given picture of old car or building, to virtually renew the object in the image as if it was new? Runa Takahashi from Yokohama National University, Japan
showed that this can be acheived thanks to neural networks (especially
GANs), by removing dust and stains form the object, reinforcing the
color saturation, and increasing slightly the gloss contrast. Convincing
results were shown.
Visual Attributes
- Sparkle: It is possible to record the dynamic light
signal at the origin of the sensation of sparckle into one image? A
clever idea in this direction has been defended by Shuuhei Watanabe from Ricoh and Chiba University, Japan.
A line scan spectral camera captures a spectral image of the surface
while a collimated light source is rotating around it: each pixel column
in the image therefore corresponds to one illumination direction.
Statistical properties of the image are then analyzed and correlation
with visual sparkle assessment is performed.
- Translucency: How do caustics help observers to assess the sensation of translucency? Davit Gigilashvili from NTNU, Norway, laureate of the Best student paper award of the conference, simulated various traslucent objects with and without causics and showed that the caustics
clearly provide information on the object’s translucency and their
presence increase the sensation of translucency in comparison to the
case where they are not visible. Jon Hardeberg, NTNU, Norway, (invited)
spoke on, "From color and spectral reproduction to appearance, BRDF, and beyond."
Materials
- A pine resin coating on wood, used in ancien Norvegian culture, has the interesting property to imitate gold. The question addressed by Oleksii Sidorov from NTNU, Norway, is the ageing of these objects. BRDF measurements were performed and compared on samples artificially aged.
- The use of AI technics for human skin is definitely a
growing research domain. Assessing the color of skin into color pictures
is a crucial issue for the e-market of cosmetic products, but remains a
challenge in absence of color calibration of the camera. When a color
target is displayed in the fireld of the camera, color calibration is
possible, but it is known that its performence is better if the color of
lighting is known. Robin Kips from L’Oréal, France,
shows that the color temperature of the lighting can be estimated by
using neural networks, which sensibly increases the color calibration
accuracy, therefore the one of the skin color estimation. Regarding skin
health, Ahmed Mohammed from NTNU, Norway, explains how neural network can improve skin oxygenation analysis from multispectral images.
- Paint: For who wants to develop prediction models for
the spectral reflectance or the color of paint mixtures, the challenge
is often to make good, uniform samples where the amounts of the
different primary paints is precisely known. Vahid Babaei et Azadeh A. Shahmirzadi from Max Planck Institute, Germany,
produced two color charts based on oil paints and watercolor and
created a free database, given as supplemetary material of their paper.
- 3D printing: Ingeborg Tastl from HP Labs, USA,
presented the recent progresses and remaining challenges in material
appearance reproduction by 3D printing, and exposed a 3D color gamut.
Conference Program
Monday January 27, 2020
Material Appearance 2020 Conference Introduction
Session Chair:
Mathieu Hebert, Université Jean Monnet de Saint Etienne (France)
9:20 – 9:30 AM
Regency C
KEYNOTE: 3D Digitization and Optical Material Interactions
Session Chair: Ingeborg Tastl, HP Labs, HP Inc. (United States)
9:30 – 10:10 AM
Regency C
MAAP-020
Capturing and 3D rendering of optical behavior: The physical approach to realism, Martin Ritz, Fraunhofer Institute for Computer Graphics Research (Germany)
Martin Ritz has been deputy head of Competence Center Cultural Heritage Digitization with the Fraunhofer Institute for Computer Graphics Research IGD since 2012, after working as research fellow for three years at Fraunhofer IGD in the department of Industrial Applications (today: Interactive Engineering Technologies). In parallel to technical coordination, his research topics include acquisition of 3D geometry as well as optical material properties, meaning light interaction of surfaces up to complete objects, for arbitrary combinations of light and observer directions. Challenges in both domains are equally design and implementation of algorithms as well as conceptualization and realization of novel scanning systems in hardware and software with the goal of complete automation in mind. Martin Ritz received his MS in Informatics (2009) from the Technische Universität Darmstadt. The focus of his final thesis in the domain of photogrammetry was the extension of "Multi-view Stereo" by the advantages of the "Photometric Stereo" approach in order to reach better results and a more complete measurement data coverage during 3D reconstruction. During his studies at the University of Colorado at Boulder, (United States), he received his MS in computer science (2008). His bachelor of science thesis from 2006 in the context of the European research project SmartSketches at Fraunhofer IGD targeted the implementation of a consistency mechanism satisfying continuity constraints between freeform surfaces.
10:10 – 10:50 AM Coffee Break
Sparkle, Gloss, Texture, and Translucency
Session Chair:
Mathieu Hebert, Université Jean Monnet de Saint Etienne (France)
10:50 AM – 12:10 PM
Regency C
10:50MAAP-030
One-shot multi-angle measurement device for evaluating the sparkle impression (JIST-first), Shuhei Watanabe, Ricoh Company, Ltd. (Japan)
11:10MAAP-031
Appearance reproduction of material surface with strong specular reflection, Shoji Tominaga1,2, Giuseppe Guarnera2, and Norihiro Tanaka1; 1Nagano University (Japan) and 2Norwegian University of Science and Technology (Norway)
11:30MAAP-032
BTF image recovery based on U-Net and texture interpolation, Naoki Tada and Keita Hirai, Chiba University (Japan)
11:50MAAP-033
Caustics and translucency perception, Davit Gigilashvili, Lucas Dubouchet, Jon Yngve Hardeberg, and Marius Pedersen, Norwegian University of Science and Technology (Norway)
DISCUSSION; Material Appearance Morning Q&A
Session Chairs:
Mathieu Hebert, Université Jean Monnet de Saint Etienne (France) and Ingeborg Tastl, HP Labs, HP Inc. (United States)
12:10 – 12:30 PM
Regency C
12:30 – 2:00 PM Lunch
PLENARY: Frontiers in Computational Imaging
Session Chairs: Jonathan Phillips, Google Inc. (United States) and Radka Tezaur, Intel Corporation (United States)
2:00 – 3:10 PM
Grand Peninsula D
Imaging the unseen: Taking the first picture of a black hole, Katherine Bouman, California Institute of Technology (United States)
Katherine Bouman is an assistant professor in the Computing and Mathematical Sciences Department at the California Institute of Technology. Before joining Caltech, she was a postdoctoral fellow in the Harvard-Smithsonian Center for Astrophysics. She received her PhD in the Computer Science and Artificial Intelligence Laboratory (CSAIL) at MIT in EECS. Before coming to MIT, she received her bachelor's degree in electrical engineering from the University of Michigan. The focus of her research is on using emerging computational methods to push the boundaries of interdisciplinary imaging.
3:10 – 3:30 PM Coffee Break
Aging and Renewing
Session Chair:
Shoji Tominaga, Chiba University (Japan)
3:30 – 4:10 PM
Regency C
3:30MAAP-060
Changes in the visual appearance of polychrome wood caused by (accelerated) aging, Oleksii Sidorov1, Jon Yngve Hardeberg1, Sony George1, Joshua Harvey2, and Hannah Smithson2; 1Norwegian University of Science and Technology (Norway) and 2University of Oxford (United Kingdom)
3:50MAAP-061
Image processing method for renewing old objects using deep learning, Runa Takahashi and Katsunori Okajima, Yokohama National University (Japan)
DISCUSSION: Material Appearance Afternoon Q&A
Session Chairs:
Mathieu Hebert, Université Jean Monnet de Saint Etienne (France) and Ingeborg Tastl, HP Labs, HP Inc. (United States)
4:10 – 4:30 PM
Regency C
5:00 – 6:00 PM All-Conference Welcome Reception
Tuesday January 28, 2020
7:30 – 8:45 AM Women in Electronic Imaging Breakfast (pre-registration required)
Skin and Deep Learning
Session Chair:
Gabriel Marcu, Apple Inc. (United States)
8:45 – 9:30 AM
Regency C
This session is jointly sponsored by: Color Imaging XXV: Displaying, Processing, Hardcopy, and Applications, and Material Appearance 2020.
8:45
Conference Welcome
8:50MAAP-082
Beyond color correction: Skin color estimation in the wild through deep learning, Robin Kips, Quoc Tran, Emmanuel Malherbe, and Matthieu Perrot, L'Oréal Research and Innovation (France)
9:10COLOR-083
SpectraNet: A deep model for skin oxygenation measurement from multi-spectral data, Ahmed Mohammed, Mohib Ullah, and Jacob Bauer, Norwegian University of Science and Technology (Norway)
Spectral Dataset
Session Chair:
Ingeborg Tastl, HP Labs, HP Inc. (United States)
9:30 – 10:10 AM
Regency C
This session is jointly sponsored by: Color Imaging XXV: Displaying, Processing, Hardcopy, and Applications, and Material Appearance 2020.
9:30MAAP-106
Visible to near infrared reflectance hyperspectral images dataset for image sensors design, Axel Clouet1, Jérôme Vaillant1, and Célia Viola2; 1CEA-LETI and 2CEA-LITEN (France)
9:50MAAP-107
A multispectral dataset of oil and watercolor paints, Vahid Babaei1, Azadeh Asadi Shahmirzadi2, and Hans-Peter Seidel1; 1Max-Planck-Institut für Informatik and 2Consultant (Germany)
10:00 AM – 7:30 PM Industry Exhibition - Tuesday
10:10 – 10:40 AM Coffee Break
Color and Appearance Reproduction
Session Chair:
Mathieu Hebert, Université Jean Monnet de Saint Etienne (France)
10:40 AM – 12:30 PM
Regency C
This session is jointly sponsored by: Color Imaging XXV: Displaying, Processing, Hardcopy, and Applications, and Material Appearance 2020.
10:40MAAP-396
From color and spectral reproduction to appearance, BRDF, and beyond, Jon Yngve Hardeberg, Norwegian University of Science and Technology (NTNU) (Norway)
11:10MAAP-120
HP 3D color gamut – A reference system for HP’s Jet Fusion 580 color 3D printers, Ingeborg Tastl1 and Alexandra Ju2; 1HP Labs, HP Inc. and 2HP Inc. (United States)
11:30COLOR-121
Spectral reproduction: Drivers, use cases, and workflow, Tanzima Habib, Phil Green, and Peter Nussbaum, Norwegian University of Science and Technology (Norway)
11:50COLOR-122
Parameter estimation of PuRet algorithm for managing appearance of material objects on display devices (JIST-first), Midori Tanaka, Ryusuke Arai, and Takahiko Horiuchi, Chiba University (Japan)
12:10COLOR-123
Colorimetrical performance estimation of a reference hyperspectral microscope for color tissue slides assessment, Paul Lemaillet and Wei-Chung Cheng, US Food and Drug Administration (United States)
12:30 – 2:00 PM Lunch
PLENARY: Automotive Imaging
Session Chairs: Jonathan Phillips, Google Inc. (United States) and Radka Tezaur, Intel Corporation (United States)
2:00 – 3:10 PM
Grand Peninsula D
Imaging in the autonomous vehicle revolution, Gary Hicok, NVIDIA Corporation (United States)
Gary Hicok is senior vice president of hardware development at NVIDIA, and is responsible for Tegra System Engineering, which oversees Shield, Jetson, and DRIVE platforms. Prior to this role, Hicok served as senior vice president of NVIDIA’s Mobile Business Unit. This vertical focused on NVIDIA’s Tegra mobile processor, which was used to power next-generation mobile devices as well as in-car safety and infotainment systems. Before that, Hicok ran NVIDIA’s Core Logic (MCP) Business Unit also as senior vice president. Throughout his tenure with NVIDIA, Hicok has also held a variety of management roles since joining the company in 1999, with responsibilities focused on console gaming and chipset engineering. He holds a BSEE from Arizona State University and has authored 33 issued patents.
3:10 – 3:30 PM Coffee Break
5:30 – 7:30 PM Symposium Demonstration Session