One of the most striking characteristics of human behavior in contrast to all other animal is that we show extraordinary variability across populations. Human cultural diversity is a biological oddity. More specifically, we propose that what makes humans unique is the nature of the individual ontogenetic process, that results in this unparalleled cultural diversity. Hence, our central question is: How is human ontogeny adapted to cultural diversity and how does it contribute to it? This question is critical, because cultural diversity does not only entail our predominant mode of adaptation to local ecologies, but is key in the construction of our cognitive architecture. The colors we see, the tones that we hear, the memories we form, the norms we adhere to are all the consequence of an interaction between our emerging cognitive system and our lived experiences. While psychologists make careers measuring cognitive systems, we are terrible at measuring experience. The standard methods all face unsurmountable limitations. In our department, we hope to apply Machine Learning, Deep Learning and Computer Vision to automatically extract developmentally important indicators of humans’ daily experience. Similarly to the way that modern sequencing technologies allow us to study the human genotype at scale, applying AI methods to reliably quantify humans’ lived experience would allow us to study the human behavioral phenotype at scale, and fundamentally alter the science of human behavior and its application in education, mental health and medicine: The phenotyping revolution.
Organizers: Timo Bolkart
Imagine a futuristic version of Google Street View that could dial up any possible place in the world, at any possible time. Effectively, such a service would be a recording of the plenoptic function—the hypothetical function described by Adelson and Bergen that captures all light rays passing through space at all times. While the plenoptic function is completely impractical to capture in its totality, every photo ever taken represents a sample of this function. I will present recent methods we've developed to reconstruct the plenoptic function from sparse space-time samples of photos—including Street View itself, as well as tourist photos of famous landmarks. The results of this work include the ability to take a single photo and synthesize a full dawn-to-dusk timelapse video, as well as compelling 4D view synthesis capabilities where a scene can simultaneously be explored in space and time.
Game Development requires a vast array of tools, techniques, and expertise, ranging from game design, artistic content creation, to data management and low level engine programming. Yet all of these domains have one kind of task in common - the transformation of one kind of data into another. Meanwhile, advances in Machine Learning have resulted in a fundamental change in how we think about these kinds of data transformations - allowing for accurate and scalable function approximation, and the ability to train such approximations on virtually unlimited amounts of data. In this talk I will present how these two fundamental changes in Computer Science affect game development - how they can be used to improve game technology as well as the way games are built - and the exciting new possibilities and challenges they bring along the way.
Organizers: Abhinanda Ranjit Punnakkal
I will present three recent projects within the 3D Deep Learning research line from my team at Google Research: (1) a deep network for reconstructing the 3D shape of multiple objects appearing in a single RGB image (ECCV'20). (2) a new conditioning scheme for normalizing flow models. It enables several applications such as reconstructing an object's 3D point cloud from an image, or the converse problem of rendering an image given a 3D point cloud, both within the same modeling framework (CVPR'20); (3) a neural rendering framework that maps a voxelized object into a high quality image. It renders highly-textured objects and illumination effects such as reflections and shadows realistically. It allows controllable rendering: geometric and appearance modifications in the input are accurately represented in the final rendering (CVPR'20).
Babies learn with very little supervision, and, even when supervision is present, it comes in the form of an unknown spoken language that also needs to be learned. How can kids make sense of the world? In this work, I will show that an agent that has access to multimodal data (like vision, audition or touch) can use the correlation between images and sounds to discover objects in the world without supervision. I will show that ambient sounds can be used as a supervisory signal for learning to see and vice versa (the sound of crashing waves, the roar of fast-moving cars – sound conveys important information about the objects in our surroundings). I will describe an approach that learns, by watching videos without annotations, to locate image regions that produce sounds, and to separate the input sounds into a set of components that represents the sound from each pixel. I will also discuss our recent work on capturing tactile information.
Organizers: Arjun Chandrasekaran
Learning the embedding space, where semantically similar objects are located close together and dissimilar objects far apart, is a cornerstone of many computer vision applications. Existing approaches usually learn a single metric in the embedding space for all available data points,which may have a very complex non-uniform distribution with different notions of similarity between objects, e.g. appearance, shape, color or semantic meaning. We approach this problem by using the embedding space more efficiently by jointly splitting the embedding space and data into K smaller sub-problems. It divides both, the data and the embedding space into K subsets and learns K separate distance metrics in the non-overlapping subspaces of the embedding space, defined by groups of neurons in the embedding layer of the neural network. In the second part of the talk, we show that, at least for proximal animal classes such as chimpanzees, it is possible to transfer the knowledge existing in dense pose recognition for humans, as well as inmore general object detectors and segmenters, to the problem of dense pose recognition in other classes. We do this by (1) establishing a DensePose model for the new animal which is also geometrically aligned to humans (2) introducing a multi-head R-CNN architecture that facilitates transfer of multiple recognition tasks between classes, (3) finding which combination of known classes can be transferred most effectively to the new animal and (4) using self-calibrated uncertainty heads to generate pseudo-labels graded by quality for training a model for this class.
Organizers: Nikos Athanasiou
In recent years, commodity 3D sensors have become widely available, spawning significant interest in both offline and real-time 3D reconstruction. While state-of-the-art reconstruction results from commodity RGB-D sensors are visually appealing, they are far from usable in practical computer graphics applications since they do not match the high quality of artist-modeled 3D graphics content. One of the biggest challenges in this context is that obtained 3D scans suffer from occlusions, thus resulting in incomplete 3D models. In this talk, I will present a data-driven approach towards generating high quality 3D models from commodity scan data, and the use of these geometrically complete 3D models towards semantic and texture understanding of real-world environments.
Organizers: Yinghao Huang
How can we tell that a video is playing backwards? People's motions look wrong when the video is played backwards--can we develop an algorithm to distinguish forward from backward video? Similarly, can we tell if a video is sped-up? We have developed algorithms to distinguish forwards from backwards video, and fast from slow. Training algorithms for these tasks provides a self-supervised task that facilitates human activity recognition. We'll show these results, and applications of these unsupervised video learning tasks, including a method to change the timing of people in videos.
Organizers: Yinghao Huang
Applying data-driven approaches to non-rigid 3D reconstruction has been difficult, which we believe can be attributed to the lack of a large-scale training corpus. One recent approach proposes self-supervision based on non-rigid reconstruction. Unfortunately, this method fails for important cases such as highly non-rigid deformations. We first address this problem of lack of data by introducing a novel semi-supervised strategy to obtain dense interframe correspondences from a sparse set of annotations. This way, we obtain a large dataset of 400 scenes, over 390,000 RGB-D frames, and 2,537 densely aligned frame pairs; in addition, we provide a test set along with several metrics for evaluation. Based on this corpus, we introduce a data-driven non-rigid feature matching approach, which we integrate into an optimization-based reconstruction pipeline. Here, we propose a new neural network that operates on RGB-D frames, while maintaining robustness under large non-rigid deformations and producing accurate predictions. Our approach significantly outperforms both existing non-rigid reconstruction methods that do not use learned data terms, as well as learning-based approaches that only use self-supervision.
Organizers: Vassilis Choutas
In our recent work, XNect, we propose a real-time solution for the challenging task of multi-person 3D human pose estimation from a single RGB camera. To achieve real-time performance without compromising on accuracy, our approach relies on a new efficient Convolutional Neural Network architecture, and a multi-staged pose formulation. The CNN architecture is approx. 1.3x faster than ResNet-50, while achieving the same accuracy on various tasks, and the benefits extend beyond inference speed to a much smaller training memory footprint and a much higher training throughput. The proposed pose formulation jointly reasons about all the subjects in the scene, ensuring that pose inference can be done in real time even with a large number of subjects in the scene. The key insight behind the accuracy of the formulation is to split the reasoning about human pose into two distinct stages. The first stage, which is fully convolutional, infers 2D and 3D pose of body parts supported by image evidence, and reasons jointly about all subjects. The second stage, which is a small fully connected network, operates on each individual subject, and uses the context of the visibly body parts and learned pose priors, to infer the 3D pose of the missing body parts. A third stage on top reconciles the 2D and 3D poses per frame and across time, to produce a temporally stable kinematic skeleton. In this talk, we will briefly discuss the proposed Convolutional Neural Network architecture and the possible benefits it might bring to your workflow. The other part of the talk would be on how the pose formulation proposed in this work came to be, what its advantages are, and how it can be extended to other related problems.
Organizers: Yinghao Huang