I am a Researcher at Apple AI Research. My research interest lies on the intersection of geometry and deep learning for 3D scene understanding. This includes 3D reconstruction, depth estimation, optical flow, 3D representation learning and using geometry for self-supervision.
Faces, their shape, and their motion are essential to communication. Consequently, we want a model of the face that can capture the full range of face shapes and expressions. Such a model should be realistic, easy to animate, easy to fit to data, and should support inference about human emotion and speech....
Much of our work focuses on 3D models of objects and scenes. We would like to take advantage of current deep learning approaches in representing and reasoning about 3D. Unfortunately, the standard 2D convolutional models do not readily extend to 3D because they do not match well current 3D data structures....
Deep learning has brought rapid progress for many computer vision problems but current methods require large training datasets with annotated ground truth. Human annotators tend to be reasonably efficient for tasks like sparse 2D joint estimation, however annotation for other tasks like dense optical...
We view optical flow as the projection of the 3D motion field into the image plane. Until recently, optical flow algorithms were designed by hand and incorporated various heuristics. Deep learning methods provide an opportunity to move away from hand-crafted models but have several limitations. The key one...
Understanding human action requires modeling and understanding human movement. While we mostly focus on 3D human movement, what is directly observable in videos is the 2D optical flow. Previous work has shown that flow is useful for action recognition and, consequently, we explore how to better estimate hu...
Historically, optical flow methods make generic, spatially homogeneous, assumptions about the spatial structure of the 2D image motion. In reality, optical flow varies across an image depending on object class. Simply put, different objects move differently. For rigid objects, the motion is related to the ...
Learning to solve optical flow in an end-to-end fashion from examples is attractive as deep neural networks allow for learning more complex hierarchical flow representations directly from annotated data. However, training such models requires large datasets and obtaining ground truth for real images is challenging as labeling dense ...
In International Conference on 3D Vision (3DV), November 2020 (inproceedings)
Photo-realistic visualization and animation of expressive human faces have been a long standing challenge. 3D face modeling methods provide parametric control but generates unrealistic images, on the other hand, generative 2D models like GANs (Generative Adversarial Networks) output photo-realistic face images, but lack explicit control. Recent methods gain partial control, either by attempting to disentangle different factors in an unsupervised manner, or by adding control post hoc to a pre-trained model. Unconditional GANs, however, may entangle factors that are hard to undo later. We condition our generative model on pre-defined control parameters to encourage disentanglement in the generation process. Specifically, we condition StyleGAN2 on FLAME, a generative 3D face model.
While conditioning on FLAME parameters yields unsatisfactory results, we find that conditioning on rendered FLAME geometry and photometric details works well. This gives us a generative 2D face model named GIF (Generative Interpretable Faces) that offers FLAME's parametric control. Here, interpretable refers to the semantic meaning of different parameters. Given FLAME parameters for shape, pose, expressions, parameters for appearance, lighting, and an additional style vector, GIF outputs photo-realistic face images. We perform an AMT based perceptual study to quantitatively and qualitatively evaluate how well GIF follows its conditioning. The code, data, and trained model are publicly available for research purposes at http://gif.is.tue.mpg.de
In Computer Vision and Pattern Recognition (CVPR), pages: 6468-6477, IEEE, June 2020 (inproceedings)
Three-dimensional human body models are widely used in the analysis of human pose and motion. Existing models, however, are learned from minimally-clothed 3D scans and thus do not generalize to the complexity of dressed people in common images and videos. Additionally, current models lack the expressive power needed to represent the complex non-linear geometry of pose-dependent clothing shape. To address this, we learn a generative 3D mesh model of clothed people from 3D scans with varying pose and clothing. Specifically, we train a conditional Mesh-VAE-GAN to learn the clothing deformation from the SMPL body model, making clothing an additional term on SMPL. Our model is conditioned on both pose and clothing type, giving the ability to draw samples of clothing to dress different body shapes in a variety of styles and poses. To preserve wrinkle detail, our Mesh-VAE-GAN extends patchwise discriminators to 3D meshes. Our model, named CAPE, represents global shape and fine local structure, effectively extending the SMPL body model to clothing. To our knowledge, this is the first generative model that directly dresses 3D human body meshes and generalizes to different poses.
International Journal of Computer Vision (IJCV), (128):873-890, April 2020 (article)
The optical flow of humans is well known to be useful for the analysis of human action. Recent optical flow methods focus on training deep networks to approach the problem. However, the training data used by them does not cover the domain of human motion. Therefore, we develop a dataset of multi-human optical flow and train optical flow networks on this dataset. We use a 3D model of the human body and motion capture data to synthesize realistic flow fields in both single-and multi-person images. We then train optical flow networks to estimate human flow fields from pairs of images. We demonstrate that our trained networks are more accurate than a wide range of top methods on held-out test data and that they can generalize well to real image sequences. The code, trained models and the dataset are available for research.
The motion of the world is inherently dependent on the spatial structure of the world and its geometry. Therefore, classical optical flow methods try to model this geometry to solve for the motion. However, recent deep learning methods take a completely different approach. They try to predict optical flow by learning from labelled data. Although deep networks have shown state-of-the-art performance on classification problems in computer vision, they have not been as effective in solving optical flow. The key reason is that deep learning methods do not explicitly model the structure of the world in a neural network, and instead expect the network to learn about the structure from data. We hypothesize that it is difficult for a network to learn about motion without any constraint on the structure of the world. Therefore, we explore several approaches to explicitly model the geometry of the world and its spatial structure in deep neural networks.
The spatial structure in images can be captured by representing it at multiple scales. To represent multiple scales of images in deep neural nets, we introduce a Spatial Pyramid Network (SpyNet). Such a network can leverage global information for estimating large motions and local information for estimating small motions. We show that SpyNet significantly improves over previous optical flow networks while also being the smallest and fastest neural network for motion estimation. SPyNet achieves a 97% reduction in model parameters over previous methods and is more accurate.
The spatial structure of the world extends to people and their motion. Humans have a very well-defined structure, and this information is useful in estimating optical flow for humans. To leverage this information, we create a synthetic dataset for human optical flow using a statistical human body model and motion capture sequences. We use this dataset to train deep networks and see significant improvement in the ability of the networks to estimate human optical flow.
The structure and geometry of the world affects the motion. Therefore, learning about the structure of the scene together with the motion can benefit both problems. To facilitate this, we introduce Competitive Collaboration, where several neural networks are constrained by geometry and can jointly learn about structure and motion in the scene without any labels. To this end, we show that jointly learning single view depth prediction, camera motion, optical flow and motion segmentation using Competitive Collaboration achieves state-of-the-art results among unsupervised approaches.
Our findings provide support for our hypothesis that explicit constraints on structure and geometry of the world lead to better methods for motion estimation.
In Proceedings International Conference on Computer Vision (ICCV), pages: 2404-2413, IEEE, November 2019, ISSN: 2380-7504 (inproceedings)
Deep neural nets achieve state-of-the-art performance on the problem of optical flow estimation. Since optical flow is used in several safety-critical applications like self-driving cars, it is important to gain insights into the robustness of those techniques. Recently, it has been shown that adversarial attacks easily fool deep neural networks to misclassify objects. The robustness of optical flow networks to adversarial attacks, however, has not been studied so far. In this paper, we extend adversarial patch attacks to optical flow networks and show that such attacks can compromise their performance. We show that corrupting a small patch of less than 1% of the image size can significantly affect optical flow estimates. Our attacks lead to noisy flow estimates that extend significantly beyond the region of the attack, in many cases even completely erasing the motion of objects in the scene. While networks using an encoder-decoder architecture are very sensitive to these attacks, we found that networks using a spatial pyramid architecture are less affected. We analyse the success and failure of attacking both architectures by visualizing their feature maps and comparing them to classical optical flow techniques which are robust to these attacks. We also demonstrate that such attacks are practical by placing a printed pattern into real scenes.
In Proceedings IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pages: 12240-12249, June 2019 (inproceedings)
We address the unsupervised learning of several interconnected problems in low-level vision: single view depth prediction, camera motion estimation, optical flow, and segmentation of a video into the static scene and moving regions. Our key insight is that these four fundamental vision problems are coupled through geometric constraints. Consequently, learning to solve them together simplifies the problem because the solutions can reinforce each other. We go beyond previous work by exploiting geometry more explicitly and segmenting the scene into static and moving regions. To that end, we introduce Competitive Collaboration, a framework that facilitates the coordinated training of multiple specialized neural networks to solve complex problems. Competitive Collaboration works much like expectation-maximization, but with neural networks that act as both competitors to explain pixels that correspond to static or moving regions, and as collaborators through a moderator that assigns pixels to be either static or independently moving. Our novel method integrates all these problems in a common framework and simultaneously reasons about the segmentation of the scene into moving objects and the static background, the camera motion, depth of the static scene structure, and the optical flow of moving objects. Our model is trained without any supervision and achieves state-of-the-art performance among joint unsupervised methods on all sub-problems.
In Proceedings IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pages: 10101-10111, June 2019 (inproceedings)
Audio-driven 3D facial animation has been widely explored, but achieving realistic, human-like performance is still unsolved. This is due to the lack of available 3D datasets, models, and standard evaluation metrics. To address this, we introduce a unique 4D face dataset with about 29 minutes of 4D scans captured at 60 fps and synchronized audio from 12 speakers. We then train a neural network on our dataset that factors identity from facial motion. The learned model, VOCA (Voice Operated Character Animation) takes any speech signal as input—even speech in languages other than English—and realistically animates a wide range of adult faces. Conditioning on subject labels during training allows the model to learn a variety of realistic speaking styles. VOCA also provides animator controls to alter speaking style, identity-dependent facial shape, and pose (i.e. head, jaw, and eyeball rotations) during animation. To our knowledge, VOCA is the only realistic 3D facial animation model that is readily applicable to unseen subjects without retargeting. This makes VOCA suitable for tasks like in-game video, virtual reality avatars, or any scenario in which the speaker, speech, or language is not known in advance. We make the dataset and model available for research purposes at http://voca.is.tue.mpg.de.
In European Conference on Computer Vision (ECCV), Lecture Notes in Computer Science, vol 11207, pages: 725-741, Springer, Cham, September 2018 (inproceedings)
Learned 3D representations of human faces are useful for computer vision problems such as 3D face tracking and reconstruction from images, as well as graphics applications such as character generation and animation. Traditional models learn a latent representation of a face using linear subspaces or higher-order tensor generalizations. Due to this linearity, they can not capture extreme deformations and non-linear expressions. To address this, we introduce a versatile model that learns a non-linear representation of a face using spectral convolutions on a mesh surface. We introduce mesh sampling operations that enable a hierarchical mesh representation that captures non-linear variations in shape and expression at multiple scales within the model. In a variational setting, our model samples diverse realistic 3D faces from a multivariate Gaussian distribution. Our training data consists of 20,466 meshes of extreme expressions captured over 12 different subjects. Despite limited training data, our trained model outperforms state-of-the-art face models with 50% lower reconstruction error, while using 75% fewer parameters. We also show that, replacing the expression space of an existing state-of-the-art face model with our autoencoder, achieves a lower reconstruction error. Our data, model and code are available at http://coma.is.tue.mpg.de/.
In 29th British Machine Vision Conference, September 2018 (inproceedings)
The optical flow of humans is well known to be useful for the analysis of human action. Given this, we devise an optical flow algorithm specifically for human motion and show that it is superior to generic flow methods. Designing a method by hand is impractical, so we develop a new training database of image sequences with ground truth optical flow. For this we use a 3D model of the human body and motion capture data to synthesize realistic flow fields. We then train a convolutional neural network to estimate human flow fields from pairs of images. Since many applications in human motion analysis depend on speed, and we anticipate mobile applications, we base our method on SpyNet with several modifications. We demonstrate that our trained network is more accurate than a wide range of top methods on held-out test data and that it generalizes well to real image sequences. When combined with a person detector/tracker, the approach provides a full solution to the problem of 2D human flow estimation. Both the code and the dataset are available for research.
In Proceedings IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2017, IEEE, Piscataway, NJ, USA, July 2017 (inproceedings)
We learn to compute optical flow by combining a classical spatial-pyramid formulation with deep learning. This estimates large motions in a coarse-to-fine approach by warping one image of a pair at each pyramid level by the current flow estimate and computing an update to the flow. Instead of the standard minimization of an objective function at each pyramid level, we train one deep network per level to compute the flow update. Unlike the recent FlowNet approach, the networks do not need to deal with large motions; these are dealt with by the pyramid. This has several advantages. First, our Spatial Pyramid Network (SPyNet) is much simpler and 96% smaller than FlowNet in terms of model parameters. This makes it more efficient and appropriate for embedded applications. Second, since the flow at each pyramid level is small (< 1 pixel), a convolutional approach applied to pairs of warped images is appropriate. Third, unlike FlowNet, the learned convolution filters appear similar to classical spatio-temporal filters, giving insight into the method and how to improve it. Our results are more accurate than FlowNet on most standard benchmarks, suggesting a new direction of combining classical flow methods with deep learning.
Our goal is to understand the principles of Perception, Action and Learning in autonomous systems that successfully interact with complex environments and to use this understanding to design future systems