Visual Computing Seminar (Fall 2017)
The Visual computing seminar is a weekly seminar series on topics in Visual Computing.
Why: The motivation for creating this seminar is that EPFL has a critical mass of people who are working on subtly related topics in computational photography, computer graphics, geometry processing, human–computer interaction, computer vision and signal processing. Having a weekly point of interaction will provide exposure to interesting work in this area and increase awareness of our shared interests and other commonalities like the use of similar computational tools — think of this as the visual computing edition of the “Know thy neighbor” seminar series.
Who: The target audience are faculty, students and postdocs in the visual computing disciplines, but the seminar is open to anyone and guests are welcomed. There is no need to formally enroll in a course. The format is very flexible and will include 45 minute talks with Q&A, talks by external visitors, as well as shorter presentations. In particular, the seminar is also intended as a way for students to obtain feedback on shorter ~20min talks preceding a presentation at a conference. If you are a student or postdoc in one of the visual computing disciplines, you’ll probably receive email from me soon on scheduling a presentation.
Where and when: every Wednesday in BC04 (next to the ground floor atrium). Food is served at 11:50, and the actual talk starts at 12:15.
How to be notified: If you want to be kept up to date with announcements, please send me an email and I’ll put you on the list. If you are working in LCAV, CVLAB, IVRL, LGG, LSP, IIG, CHILI, LDM or RGL, you are automatically subscribed to future announcements, so there is nothing you need to do.
You may add the seminar events to Google Calendar (click the '+' button in the bottom-right corner), or download the iCal file.
Title: Rendering Specular Microstructure Using Adaptive Gaussian Processes
Abstract: Many materials in our world are characterized by complex angularly dependent reflectance behavior that is challenging to reproduce in realistic light transport simulations. In particular, materials such as brushed metals, anodized aluminum and bumpy plastics are interesting since their surfaces produce glints, that is, strong reflections due to very high-frequency surface normal variations. Although measurements can be used to capture reflectance values for many combinations of lighting and observation angles, achieving sufficient precision requires large amounts of data, which is then difficult to use in a practical rendering scenario.
However, the characteristics of the surface and its fabrication process imply some statistical regularity in the observations, which we wish to characterize and exploit. We propose a new stochastic microfacet model able to capture the complex specular behavior of rough surfaces under sharp lighting. We model microsurface detail as a Gaussian Process, which is efficiently sampled. Users may specify desired surface statistics through an autocorrelation function, extending the class of supported surfaces compared to previous work. Our model is fully procedural and thus does not rely on any provided microsurface. Evaluation is accelerated through pruning strategies derived from the predictive distribution of slopes. Finally, we provide a proof-of-concept implementation in a rendering system to validate our method's properties experimentally.
Title: Frequency-Space Methods for Modeling Layered Materials
Abstract: Today's rendering systems use a number of bidirectional scattering distribution functions (BSDFs) representing different idealized materials such as metals or dielectrics. These produce physically plausible results, but usually fail to reproduce the complex scattering behavior present in most real world materials.
In this talk, we discuss a special subset of materials that can be described by stacked layers of existing scattering models. Correctly combining models is a challenging problem due to the difficult internal scattering that happens between the layers. We present a framework for generating layered BSDFs stored in a frequency-space representation that facilitates the adding process in a precomputation step and additionally allows for efficient evaluation and importance sampling in a rendering system. By using layers of commonly used anisotropic microfacet BSDFs as well as scattering and absorbing media as building blocks, we can substantially enlarge the space of interesting materials for rendering applications.
Title: Computational Design of Wind-up Toys
Abstract: Wind-up toys are mechanical assemblies that perform intriguing motions driven by a simple spring motor. Due to the limited motor force and small body size, wind-up toys often employ higher pair joints of less frictional contacts and connector parts of nontrivial shapes to transfer motions. These unique characteristics make them hard to design and fabricate as compared to other automata. This talk presents a computational system to aid the design of wind-up toys, focusing on constructing a compact internal wind-up mechanism to realize user-requested part motions. We use our system to design wind-up toys of various forms, fabricate a number of them using 3D printing, and show the functionality of various results.
Title: A brief history of superpixels
Abstract: Superpixels are roughly equally sized clusters of similar looking pixels. Creating superpixels reduces the number of entities in an image from a few million pixels to a few thousand clusters of pixels. This in turn reduces the computational burden of subsequent algorithms, such as those that use graphs because there are fewer edges to account for. This talk presents three superpixel segmentation algorithms. The first one is Simple Linear Iterative Clustering (SLIC), the second is a variant of this called Simple Non-iterative Clustering (SNIC), and the third is an algorithm that relaxes the requirement of having roughly equal size.
Title: Functional Stochastic Maximum Likelihood for Array Signal Processing
Abstract: Stochastic Maximum Likelihood (SML) is a popular direction of arrival (DOA) estimation technique in array signal processing. It is a parametric method that jointly estimates signal and instrument noise by maximum likelihood, achieving excellent statistical performance. Some drawbacks are the computational overhead as well as the too restrictive point-source data model, which assumes fewer sources than sensors. In this work, we propose Functional Stochastic Maximum Likelihood (FSML). It uses a general functional data model, allowing an unrestricted number of arbitrarily-shaped sources to be recovered. To this end, we leverage functional analysis tools and express the data in terms of an infinite-dimensional sampling operator acting on a Gaussian random function. We show that FSML is computationally more efficient than traditional SML, resilient to noise, and results in much better accuracy than spectral-based methods.
Title: Online Generative Model Personalization for Hand Tracking
Abstract: We present a new algorithm for real-time hand tracking on commodity depth sensing devices. Our method does not require a user-specific calibration session, but rather learns the geometry as the user performs live in front of the camera, thus enabling seamless virtual interaction at the consumer level. The key novelty in our approach is an online optimization algorithm that jointly estimates pose and shape in each frame, and determines the uncertainty in such estimates. This knowledge allows the algorithm to integrate per-frame estimates over time, and build a personalized geometric model of the captured user. Our approach can easily be integrated in state-of-the art continuous generative motion tracking software. We provide a detailed evaluation that shows how our approach achieves accurate motion tracking for real-time applications, while significantly simplifying the workflow of accurate hand performance capture.
Title: Statistically-Motivated Second-Order Pooling
Abstract: Second-order pooling, a.k.a.~bilinear pooling, has proven effective for visual recognition. The recent progress in this area has focused on either designing normalization techniques for second-order models, or compressing the second-order representations. However, these two directions have typically been followed separately, and without any clear statistical motivation. Here, by contrast, we introduce a statistically-motivated framework that jointly tackles normalization and compression of second-order representations. To this end, we design a parametric vectorization layer, which maps a covariance matrix, known to follow a Wishart distribution, to a vector whose elements can be shown to follow a Chi-square distribution. We then propose to make use of a square-root normalization, which makes the distribution of the resulting representation converge to a Gaussian, thus complying with the standard machine learning assumption. As evidenced by our experiments, this lets us outperform the state-of-the-art second-order models on several benchmark recognition datasets.
Title: Computational Design of Programmable Auxetic Materials
Abstract: We present a computational method for designing programmable auxetic materials, i.e., flat materials with spatially varying physical properties optimized to approximate one target 3D surface when actuated. A key property of our approach is that the 3D surface is encoded in the 2D pattern. Leveraging the theory of conformal geometry, we control the maximal local scaling of the material to best match the surface deformation required to reach the target shape. The target surface is then achieved by maximally stretching the material everywhere. The required deformation can be physically realized in a simple manner without the need of any external guide surface, mold or scaffolding to define the final shape. In particular, we demonstrate that we can deform the fabricated material from its rest configuration towards the target by simply applying gravity or using an inflation approach. Finally, we present results that highlight potential applications in a wide spectrum of domains, ranging from small-scale heart stents to large-scale air-supported domes.
Title: Learning Active Learning from Data
Abstract: In this work, we suggest a novel data-driven approach to active learning (AL). The key idea is to train a regressor that predicts the expected error reduction for a candidate sample in a particular learning state. By formulating the query selection procedure as a regression problem we are not restricted to working with existing AL heuristics; instead, we learn strategies based on experience from previous AL outcomes. We show that a strategy can be learnt either from simple synthetic 2D datasets or from a subset of domain-specific data. Our method yields strategies that work well on real data from a wide range of domains.