Computational Imaging and Analysis of Scene Appearance
Lead Research Organisation:
Imperial College London
Department Name: Computing
Abstract
Appearance modeling has a rich history in the fields of optics, computer graphics and vision. The emergent cross-disciplinary area of computational photography has led to recent advances in measurement based appearance modeling. This fellowship project aims to further extend the boundaries of appearance modeling by investigating novel techniques in computational imaging for measurement of scene appearance and its analysis for novel graphics, vision and imaging applications.
The goal is to advance the state of the art in appearance measurements in controlled conditions as well as detailed appearance modeling in uncontrolled environments. While traditional reflectometry techniques have mostly accounted for geometric optics, this project proposes novel imaging techniques that extensively consider wave properties of polarization and spectra of incident and reflected light. The project targets novel applications of such imaging in estimation of surface reflectance of complex materials, photometric reconstruction of transparent objects, and measurement of layered scattering in translucent materials. Besides having significant applications in graphics and vision, the proposed approach has broader applications in applied optics and medical imaging.
The project proposes to further generalize such appearance measurements to uncontrolled outdoor settings in this ambitious endeavor. This will include measurement and modeling of sky polarization fields and spectral modeling of natural scene reflectance. Furthermore, the project proposes novel statistical analysis of such imagery including both spatial and angular domain appearance statistics for appearance estimation from sparse uncontrolled measurements. This has varied applications in acquisition and rendering of outdoor scenes for graphics and virtual reality applications, and in large scale scene recognition and analysis for vision, robotics and remote sensing applications.
The goal is to advance the state of the art in appearance measurements in controlled conditions as well as detailed appearance modeling in uncontrolled environments. While traditional reflectometry techniques have mostly accounted for geometric optics, this project proposes novel imaging techniques that extensively consider wave properties of polarization and spectra of incident and reflected light. The project targets novel applications of such imaging in estimation of surface reflectance of complex materials, photometric reconstruction of transparent objects, and measurement of layered scattering in translucent materials. Besides having significant applications in graphics and vision, the proposed approach has broader applications in applied optics and medical imaging.
The project proposes to further generalize such appearance measurements to uncontrolled outdoor settings in this ambitious endeavor. This will include measurement and modeling of sky polarization fields and spectral modeling of natural scene reflectance. Furthermore, the project proposes novel statistical analysis of such imagery including both spatial and angular domain appearance statistics for appearance estimation from sparse uncontrolled measurements. This has varied applications in acquisition and rendering of outdoor scenes for graphics and virtual reality applications, and in large scale scene recognition and analysis for vision, robotics and remote sensing applications.
Planned Impact
The proposed fellowship project will extend the boundaries of measurement based appearance modeling in both controlled laboratory conditions as well as in uncontrolled outdoor environments and hence has many applications in graphics, vision and applied optics. A major application area is digitization of shape and reflectance of material samples for realistic modeling of digital props in the entertainment industry for visual effects and video games. The proposed computational imaging techniques for appearance acquisition also have the potential to benefit applications outside the entertainment industry in areas such as imaging and digitization of cultural heritage artifacts in museums and art galleries for posterity, surface and subsurface imaging in diagnostic medical imaging. There are other possible applications in computer aided manufacturing for visualization and analysis/inspection of surface appearance of manufactured materials. The proposed reflectometry approaches for outdoor environments have the potential to impact the realism and quality of virtual reality and scene visualization systems and consumer applications such as Google StreetView/Earth, as well as 3D sensing and large scale scene analysis applications. Beyond graphics and vision, the study of sky polarization fields and spectral modeling of natural scenes have the potential for impact in remote sensing and atmospheric studies.
Industrial impact will be made with active engagement with industrial partners and connections in various domains including visual effects, games, cosmetics, architecture, etc. Engagement will also be done with local museums and art galleries for applications of the research in cultural heritage preservation. Applications of the research in medical imaging areas will also be explored in collaboration with respective departments at Imperial.
Academic impact will be made through research dissemination at leading venues of publications, invited talks to research groups and presentations at focused workshops on the topic of appearance acquisition and modeling.
Broader societal impact will be made in the form of public communication of the research at Imperial festivals and outreach events. A cross-disciplinary international scientific seminar on imaging will also be organized to exchange ideas between researchers from various communities.
Industrial impact will be made with active engagement with industrial partners and connections in various domains including visual effects, games, cosmetics, architecture, etc. Engagement will also be done with local museums and art galleries for applications of the research in cultural heritage preservation. Applications of the research in medical imaging areas will also be explored in collaboration with respective departments at Imperial.
Academic impact will be made through research dissemination at leading venues of publications, invited talks to research groups and presentations at focused workshops on the topic of appearance acquisition and modeling.
Broader societal impact will be made in the form of public communication of the research at Imperial festivals and outreach events. A cross-disciplinary international scientific seminar on imaging will also be organized to exchange ideas between researchers from various communities.
Organisations
- Imperial College London (Fellow, Lead Research Organisation)
- Swiss Federal Institute of Technology in Lausanne (EPFL) (Collaboration)
- University College London (Collaboration)
- LOUGHBOROUGH UNIVERSITY (Collaboration)
- College of William & Mary (Collaboration)
- Norwegian University of Science and Technology (NTNU) (Collaboration)
- UNIVERSITY OF YORK (Collaboration)
- UNIVERSITY OF BRITISH COLUMBIA (Collaboration)
People |
ORCID iD |
Abhijeet Ghosh (Principal Investigator / Fellow) |
Publications
Deschaintre V
(2020)
Guided Fine-Tuning for Large-Scale Material Transfer
in Computer Graphics Forum
Deschaintre V
(2021)
Deep Polarization Imaging for 3D Shape and SVBRDF Acquisition
Gitlina Y
(2018)
Practical measurement-based spectral rendering of human skin
Gitlina Y
(2020)
Practical Measurement and Reconstruction of Spectral Skin Reflectance
in Computer Graphics Forum
Guarnera D
(2016)
BRDF Representation and Acquisition
in Computer Graphics Forum
Henzler P
(2021)
Generative modelling of BRDF textures from flash images
in ACM Transactions on Graphics
Hu Y
(2022)
An Inverse Procedural Modeling Pipeline for SVBRDF Maps
in ACM Transactions on Graphics
Kampouris C.
(2018)
Diffuse-specular separation using binary spherical gradient illumination
Description | Research funded by this grant (fellowship) has contributed to many research projects including the study of diffraction in surface reflectance due to microscopic grating structure (published and presented at the SIGGRAPH Asia 2018 conference and another publication in ACM Transactions on Graphics with presentation at SIGGRAPH 2017), practical real-time rendering of surface diffraction (CVMP 2017 conference), the study of practical image based relighting using common room lighting (CVMP 2016 conference), the study of the texture and shape of various types of common material fabric (ECCV 2016 conference), the study of separation of surface reflectance and subsurface scattering using a combination of polarization imaging and light field photography (EGSR 2016 conference, Sensors 2018 journal paper), transmission imaging based high-quality 3D reconstruction of axially symmetric transparent objects (e.g., glasses, goblets, carafe etc, CVPR 2017 conference), practical on-site surface reflectometry in completely uncontrolled outdoor environments exploiting polarization imaging (published and presented at the SIGGRAPH Asia 2017 conference), practical on-site estimation of material reflectance (BRDF) of a homogeneous exemplar sample of arbitrary shape (published and presented at EGSR 2019 conference), novel neural encoding and compression of bidirection texture functions (BTFs), particularly representing the appearance of fabrics (published and presented at Eurographics 2019 and Eurographics 2020 conferences), and practical measurement and modeling of spectral skin reflectance (published and presented at EGSR 2020). The diffraction study has applications in life sciences and material manufacturing besides in computer graphics, while the study of fabrics and their structure and appearance has applications in robotics, fashion and advertising beyond computer vision and graphics. The study of spectral skin reflectance has applications in dermatology and cosmetics besides realistic rendering of skin for computer graphics. The EPSRC fellowship has also funded the construction of the RGI group's multispectral Lightstage which has contributed two conference papers on multiview facial capture (EGSR 2018) and deep learning based facial reconstruction from in-the-wild face images (CVPR 2020), and a journal paper on spectral skin reflectance capture (EGSR 2020). The data acquired with the device will likely lead to several follow-up publications in the future in the fields of computer graphics and vision. Another line of research combining polarization imaging and deep learning for shape and reflectance acquisition has resulted in an oral paper at CVPR 2021. |
Exploitation Route | The fabric database is already being used by other research groups to test and develop computer vision algorithms for material recognition/reconstruction. The CVMP 2016 work on practical image based relighting could be used by VFX studios for a cost-effective digital compositing solution. The EGSR 2016 and the EGSR 2020 work on light field imaging and spectral skin reflectance capture respectively have practical applications in medical imaging and cosmetics industry for separation of skin reflectance from subsurface scattering. The work on diffraction modelling and rendering is of interest to researcher in life sciences studying animal and insect skin that sometimes exhibit diffraction effects, as well as in the material manufacturing industry such as high resolution LCD panels as well as holographic materials often employed for decorations. The work on transparent object scanning has potential applications in 3D reconstruction of cellular structures with microscopy. The work on on-site reflectance capture and modeling has applications in the creative sector and for art and archeology. The neural BTF modelling work has applications in the fashion industry for efficient rendering of various types of fabrics. The facial capture data acquired with the multispectral Lightstage can be used for creation of realistic digital faces in the creative sector (movies, game, AR/VR), as well as for training AI algorithms to recognize and synthesize human faces for various applications. Several of the academic publications have led to IP (with patent applications) that are being licensed to Imperial College spin-off company Lumirithmic Ltd founded by the PI. |
Sectors | Creative Economy,Digital/Communication/Information Technologies (including Software),Education,Healthcare,Manufacturing, including Industrial Biotechology,Culture, Heritage, Museums and Collections,Retail,Other |
URL | http://wp.doc.ic.ac.uk/rgi/projects/ |
Description | The published research on "Polarization Imaging Reflectometry in the Wild" has found some interest among researchers at Sony R&D, Japan who requested some code associated with the project to internally try out the method for their applications with Sony's new polarization sensor. The RGI group's multispectral Lightstage, which has been built using funding from EPSRC fellowship, has been used for driving research in facial appearance capture of subjects. Multiple patent applications based on research outputs funded by the grant have been submitted (currently 2 US patents and 1GB patent granted) and the resulting IP has been licensed to PI's founded startup Lumirithmic Ltd which is an Imperial College spin-out. Another research project supported by the EPSRC fellowship on "Deep polarization imaging for shape and reflectance capture" has also resulted in IP which is now a granted US and UK patent and also licensed from Imperial College London to Lumirithmic Ltd as part of the spin-out. |
First Year Of Impact | 2017 |
Sector | Creative Economy,Digital/Communication/Information Technologies (including Software) |
Impact Types | Societal,Economic |
Description | (PRIME) - Predictive Rendering In Manufacture and Engineering |
Amount | € 4,140,701 (EUR) |
Funding ID | 956585 |
Organisation | European Commission |
Sector | Public |
Country | European Union (EU) |
Start | 10/2020 |
End | 09/2024 |
Description | Practical Facial Capture System |
Amount | £70,240 (GBP) |
Funding ID | EP/R511547/1 |
Organisation | Engineering and Physical Sciences Research Council (EPSRC) |
Sector | Public |
Country | United Kingdom |
Start | 01/2021 |
End | 12/2021 |
Title | Multispectral Light Stage |
Description | We have built a multispectral light stage (LED sphere) consisting of 168 RGB and color temperature controllable white (W+) lamps, respectively. The light stage is powered with off-the-shelf programmable MR16 lamps from Philips Colorkinetics: 168 iColor MR gen3 lamps for RGB illumination, and 168 iW MR gen3 lamps for color temperature controllable white (2700K - 5700K) illumination. The light stage can be employed for both controlled appearance measurements as well as realistic lighting reproduction applications. |
Type Of Material | Improvements to research infrastructure |
Year Produced | 2018 |
Provided To Others? | Yes |
Impact | We are currently employing the light stage in on-going research projects in my research group (Realistic Graphics & Imaging). The research apparatus has already contributed to two published papers in 2018 on appearance capture, one each at Eurographics Workshop on Material Appearance Modelling (http://wp.doc.ic.ac.uk/rgi/project/multispectral-light-stage/), and the Eurographics Symposium on Rendering (http://wp.doc.ic.ac.uk/rgi/project/diffuse-specular-separation-using-binary-spherical-gradient-illumination/). It has also contributed to two posters at SIGGRAPH 2019 on skin appearance (https://dl.acm.org/doi/10.1145/3306214.3338607) and facial capture (https://dl.acm.org/doi/10.1145/3306214.3338611). |
URL | http://wp.doc.ic.ac.uk/rgi/project/multispectral-light-stage/ |
Title | Fabric Database |
Description | Research funded by the EPSRC fellowship resulted in the collection of database for computer vision analysis of various types of fabrics including their texture and their surface shape variation in the form of surface normals. The datase contains about 2000 samples of different types of common fabrics ranging from cotton, denim, wool, synthetic etc. |
Type Of Material | Database/Collection of data |
Year Produced | 2016 |
Provided To Others? | Yes |
Impact | The database is available to researchers around the world to analyse the data and to develop new algorithms for computer vision, graphics and robotic applications. The original database and some initial analysis based on it for material recognition has already been published and presented at the ECCV 2016 conference. |
URL | https://ibug.doc.ic.ac.uk/resources/fabrics/ |
Title | RealfaceDB |
Description | This dataset contains patches of facial skin measurements with separated diffuse and specular reflectance and high-frequency surface normals information obtained using Lightstage based facial capture. |
Type Of Material | Database/Collection of data |
Year Produced | 2020 |
Provided To Others? | Yes |
Impact | The dataset and original CVPR 2020 paper has resulted in an extension journal paper in IEEE PAMI which is the top journal in computer vision. |
URL | https://github.com/lattas/avatarme |
Description | College of William and Mary |
Organisation | College of William & Mary |
Country | United States |
Sector | Academic/University |
PI Contribution | The project was mainly led by me and conducted in my group at Imperial College London. There was some technical input in the form of consulting provided by College of William and Mary. |
Collaborator Contribution | Dr Pieter Peers at the College of William and Mary was the collaborator on the project and provided some of input in the the form of consulting. |
Impact | This collaboration has resulted in a paper in Computer Graphics Forum in 2016 (presented at Eurographics 2016 http://wp.doc.ic.ac.uk/rgi/project/mobile-surface-reflectometry/), and another paper in Computer Graphics Forum in 2019 (presented at Eurographics Symposium on Rendering 2019 https://wp.doc.ic.ac.uk/rgi/project/on-site-example-based-material-appearance-acquisition/). |
Start Year | 2016 |
Description | Loughborough University |
Organisation | Loughborough University |
Department | Design School |
Country | United Kingdom |
Sector | Academic/University |
PI Contribution | I contributed to a State of the Art research report on BRDF representation and acquisition compiled by my collaborators at Loughborough Univerisity. |
Collaborator Contribution | My collaborators, led by Dr Mashhuda Glencross, at Loughborough University compiled a review on the state of the art of BRDF representation and acquisition (published in the Computer Graphics Forum journal) to which I contributed. |
Impact | This collaboration has led to a journal publication in Computer Graphics Forum reviewing the State of the Art in BRDF representation and acquisition. There has also been a presentation on the STAR report at the Eurographics 2016 conference and a short course on the same topic at the SIGGRAPH Asia 2016 conference. |
Start Year | 2016 |
Description | NTNU and University of York |
Organisation | Norwegian University of Science and Technology (NTNU) |
Country | Norway |
Sector | Academic/University |
PI Contribution | This is a collaboration with Dr Claudio Guarnera (NTNU and York) on spectral skin reflectance capture and modelling. The main work was conducted at Imperial College in the Realistic Graphics & Imaging group by PhD student Yuliya Gitlina and involved the RGI group's multispectral Lightstage research apparatus. |
Collaborator Contribution | Dr Guarnera provided technical consulting on the Lightstage spectral reflectance capture process. |
Impact | The collaboration has so far resulted in a poster at SIGGRAPH 2019: https://dl.acm.org/doi/10.1145/3306214.3338607, and a journal paper at EGSR 2020: https://onlinelibrary.wiley.com/doi/abs/10.1111/cgf.14055 |
Start Year | 2018 |
Description | NTNU and University of York |
Organisation | University of York |
Country | United Kingdom |
Sector | Academic/University |
PI Contribution | This is a collaboration with Dr Claudio Guarnera (NTNU and York) on spectral skin reflectance capture and modelling. The main work was conducted at Imperial College in the Realistic Graphics & Imaging group by PhD student Yuliya Gitlina and involved the RGI group's multispectral Lightstage research apparatus. |
Collaborator Contribution | Dr Guarnera provided technical consulting on the Lightstage spectral reflectance capture process. |
Impact | The collaboration has so far resulted in a poster at SIGGRAPH 2019: https://dl.acm.org/doi/10.1145/3306214.3338607, and a journal paper at EGSR 2020: https://onlinelibrary.wiley.com/doi/abs/10.1111/cgf.14055 |
Start Year | 2018 |
Description | UCL and EPFL |
Organisation | Swiss Federal Institute of Technology in Lausanne (EPFL) |
Country | Switzerland |
Sector | Public |
PI Contribution | This was a collaboration on a project led by Prof. Tim Weyrich at UCL, in collaboration with myself at Imperial, and Dr Wenzel Jakob at EPFL. I provided some technical consulting to this proect. |
Collaborator Contribution | The project was mainly led by UCL, specifically Prof. Tim Weyrich and his PhD student Gilles Rainer. Dr Jakob from EPFL also collaborated by providing technical consulting. |
Impact | This collaboration resulted in two journal papers at Eurographics 2019 (http://reality.cs.ucl.ac.uk/projects/btf/rainer19neural.html) and Eurographics 2020 (http://reality.cs.ucl.ac.uk/projects/btf/rainer2020unified.html). |
Start Year | 2017 |
Description | UCL and EPFL |
Organisation | University College London |
Country | United Kingdom |
Sector | Academic/University |
PI Contribution | This was a collaboration on a project led by Prof. Tim Weyrich at UCL, in collaboration with myself at Imperial, and Dr Wenzel Jakob at EPFL. I provided some technical consulting to this proect. |
Collaborator Contribution | The project was mainly led by UCL, specifically Prof. Tim Weyrich and his PhD student Gilles Rainer. Dr Jakob from EPFL also collaborated by providing technical consulting. |
Impact | This collaboration resulted in two journal papers at Eurographics 2019 (http://reality.cs.ucl.ac.uk/projects/btf/rainer19neural.html) and Eurographics 2020 (http://reality.cs.ucl.ac.uk/projects/btf/rainer2020unified.html). |
Start Year | 2017 |
Description | University of British Columbia |
Organisation | University of British Columbia |
Country | Canada |
Sector | Academic/University |
PI Contribution | This was a collaboration with Prof. Dinesh Pai in the Dept. of Computer Science at UBC on skin imaging. The project is mostly led by me and my group at Imperial. |
Collaborator Contribution | Prof. Pai provided some data collected by his group to assist with the research project |
Impact | The collaboration has resulted in a poster at SIGGRAPH 2018 and a journal paper at EGSR 2020. |
Start Year | 2017 |
Title | Deep polarization imaging for SVBRDF estimation |
Description | The method estimated high quality shape and reflectance of an object using a combination of polarization imaging and deep learning. |
IP Reference | US17/224,401 |
Protection | Patent / Patent application |
Year Protection Granted | 2022 |
Licensed | Yes |
Impact | Licensed by Imperial College London spin-out startup founded by PI. |
Title | Image Capture and Processing |
Description | Spectral skin chromophore estimation from RGB measurements. |
IP Reference | US16/940,107 |
Protection | Patent / Patent application |
Year Protection Granted | 2022 |
Licensed | Yes |
Impact | Licensed by Imperial College London spin-out startup founded by the PI. |
Title | Image Processing |
Description | The present invention relates to processing images to obtain separate mappings of specular and diffuse reflectance of an imaged object. The present invention also relates to processing images to obtain maps of specular and diffuse photometric normals of an imaged object, which may be used for modelling the imaged object. |
IP Reference | US patent application no. US-2021-0005015-A1 |
Protection | Patent granted |
Year Protection Granted | 2021 |
Licensed | Commercial In Confidence |
Impact | The IP related to the US patent is being licensed to an Imperial College spin-off company founded by the PI. |
Company Name | Lumirithmic Ltd |
Description | Lumirithmic was founded by the PI to commercialise methods of high quality appearance and shape capture developed in his research group at Imperial College. |
Year Established | 2020 |
Impact | Lumirithmic has developed practical products for high quality facial and object capture using two product lines, one a desktop based capture system and another phone-based capture system that significantly simplies 3D appearance capture for consumer applications. |
Website | https://www.lumirithmic.com/ |
Description | BMVC2016 Tutorial |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Postgraduate students |
Results and Impact | I delivered an invited conference tutorial on "Measurement based apperance modelling" at the BMCV2016 conference in York. The half day tutorial was attended by the conference attendees, most of whom were post-graduate students or post-doctoral researchers in the field of computer vision. The tutorial was a bit inter-disciplinary for the audience as it was delivered from the perspective of computer graphics research with concepts from applied optics and computational photography. |
Year(s) Of Engagement Activity | 2016 |
URL | http://bmvc2016.cs.york.ac.uk/ |
Description | CVCS 2020 Keynote |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Professional Practitioners |
Results and Impact | The PI delivered a keynote talk at the CVCS2020 conference: https://www.cvcs.no/conference-program/keynote-abhijeet-ghosh/ The Keynote talk presented a wide range of research in his Realistic Graphics & Imaging group on "Practical Spectral Imaging for Realistic Appearance Modeling" which covered projects on spectral imaging in controlled conditions for appearance modelling funded by EPSRC Early Career Fellowship EP/N006259/1. |
Year(s) Of Engagement Activity | 2020 |
URL | https://www.cvcs.no/conference-program/keynote-abhijeet-ghosh/ |
Description | CVMP 2019 Keynote Talk |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Professional Practitioners |
Results and Impact | The PI, Dr Ghosh gave a Keynote talk at the CVMP 2019 conference that presented a wide range of research in his Realistic Graphics & Imaging group on "Computational Imaging for Realistic Appearance Modelling" which covered projects on on-site surface reflectometry funded by EPSRC First Grant EP/M00192X/1 and projects on computation imaging in controlled conditions for appearance modelling funded by EPSRC Early Career Fellowship EP/N006259/1. |
Year(s) Of Engagement Activity | 2019 |
URL | https://www.cvmp-conference.org/2019/keynotes/ |
Description | Royal Society International Scientific Seminar on Imaging in Graphics, Vision and Beyond |
Form Of Engagement Activity | Participation in an activity, workshop or similar |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Professional Practitioners |
Results and Impact | I organized a multi-disciplinary seminar on imaging in co-operation with the Royal Society in May 2016. There were 20 invited attendee, mostly from within computer science (in sub-fields of computer graphics and computer vision), but also from some other disciplines such as geography, medical imaging and digital humanities. The purpose of the two day seminar at Chichely Hall was to discuss potential opportunities for cross-disciplinary collaborations and joint-ventures related to imaging. There was a lot of new interesting discussions on the needs of various non-computing fields and how the tools developed within computing could assist such fields. |
Year(s) Of Engagement Activity | 2016 |
URL | http://wp.doc.ic.ac.uk/rgi/event/royal-society-international-scientific-seminar-imaging-in-graphics-... |