Edith Tretschk

I am a Research Scientist at Meta Reality Labs Research in the San Francisco Bay Area.

I did my PhD in the Visual Computing and Artificial Intelligence department at the Max Planck Institute for Informatics with Christian Theobalt as my Ph.D. advisor.

Email  /  CV  /  Google Scholar  /  LinkedIn  /  Twitter

Research

My research lies at the intersection of computer graphics, computer vision, and machine learning. My work so far has covered 3D reconstruction; general, non-rigidly deforming objects; and neural rendering and neural scene representations.

Existing methods for the 4D reconstruction of general, non-rigidly deforming objects focus on novel-view synthesis and neglect correspondences. We propose SceNeRFlow to reconstruct a general, non-rigid scene in a time-consistent manner. Our dynamic-NeRF method takes multi-view RGB videos and background images from static cameras with known camera parameters as input. Like prior dynamic-NeRF methods, we use a backwards deformation model. We find non-trivial adaptations of this model necessary to handle larger motions. We show experimentally that, unlike prior work that only handles small motion, our method enables the reconstruction of studio-scale motions.

This paper introduces the first quantum auto-encoder for 3D point clouds. Our 3D-QAE approach is fully quantum, i.e. all its data processing components are designed for quantum hardware. Along with finding a suitable architecture, the core challenges in designing such a fully quantum model include 3D data normalisation and parameter optimisation, and we propose solutions for both these tasks. Experiments on simulated gate-based quantum hardware demonstrate that our method outperforms simple classical baselines, paving the way for a new research direction in 3D computer vision.

Jointly matching multiple, non-rigidly deformed 3D shapes is a challenging, NP-hard problem. A perfect matching is necessarily cycle-consistent: Following the pairwise point correspondences along several shapes must end up at the starting vertex of the original shape. This paper introduces the first quantum-hybrid approach for 3D shape multi-matching. Its iterative formulation is admissible to modern adiabatic quantum hardware and scales linearly with the total number of input shapes. On benchmark datasets, the proposed approach is on-par with classical multi-matching methods.

This survey focuses on state-of-the-art methods for dense non-rigid 3D reconstruction of various deformable objects and composite scenes from monocular videos or sets of monocular views. It reviews the fundamentals of 3D reconstruction from 2D image observations. We then start from general methods, and proceed towards techniques making stronger assumptions about the observed objects (e.g. human faces, bodies, hands, and animals). We conclude by discussing open challenges in the field and the social aspects associated with the usage of the reviewed methods.

Modern quantum annealers can find high-quality solutions to combinatorial optimisation objectives given as quadratic unconstrained binary optimisation (QUBO) problems. Unfortunately, obtaining suitable QUBO forms in computer vision remains challenging and currently requires problem-specific analytical derivations. In stark contrast to prior work, this paper proposes to learn QUBO forms from data through gradient backpropagation instead of deriving them. As a result, the solution encodings can be chosen flexibly and compactly. Furthermore, our methodology is general and virtually independent of the specifics of the target problem type.

This study investigates how qubits of modern quantum annealers (QA) such as D-Wave can be applied for generating truly random numbers. We show how a QA can be initialised and how the annealing schedule can be set so that after the annealing, thousands of truly random binary numbers are measured in parallel. We discuss the observed qubits' properties and their influence on the random number generation and consider various physical factors that influence the performance of our generator, i.e., digital-to-analogue quantisation errors, flux errors, temperature errors and spin bath polarisation.

This paper proposes a new SfT approach explaining the observations through simulation of a physically-based surface deformation model representing forces and material properties. In contrast to previous works, we utilise a differentiable physics-based simulator to regularise the surface evolution. In addition, we regress the material properties such as its bending coefficients, elasticity, stiffness, and material density. For the evaluation, we record with an RGB-D camera challenging real surfaces with various material properties and texture, exposed to physical forces. Our approach reconstructs the underlying deformations much more accurately than related methods.

We present Virtual Elastic Objects (VEOs): virtual objects that not only look like their real-world counterparts but also behave like them, even when subject to novel interactions. Achieving this presents multiple challenges: not only do objects have to be captured including the physical forces acting on them, then faithfully reconstructed and rendered, but also plausible material parameters found and simulated. The resulting method can handle objects composed of inhomogeneous material, with very different shapes, and it can simulate interactions with other virtual objects. We present our results using a newly collected dataset of 12 objects under a variety of force fields, which will be shared with the community.

This state-of-the-art report on advances in neural rendering focuses on methods that combine classical rendering principles with learned 3D scene representations, often now referred to as neural scene representations. In addition to methods that handle static scenes, we cover neural scene representations for modeling non-rigidly deforming objects and scene editing and composition. While most of these approaches are scene-specific, we also discuss techniques that generalize across object classes and can be used for generative tasks. In addition to reviewing these state-of-the-art methods, we provide an overview of fundamental concepts and definitions used in the current literature. We conclude with a discussion on open challenges and social implications.

We present Non-Rigid Neural Radiance Fields (NR-NeRF), a reconstruction and novel view synthesis approach for general non-rigid dynamic scenes. Our approach takes RGB images of a dynamic scene as input (e.g., from a monocular video recording), and creates a high-quality space-time geometry and appearance representation. We show that a single handheld consumer-grade camera is sufficient to synthesize sophisticated renderings of a dynamic scene from novel virtual camera views, e.g. a `bullet-time' video effect. Our formulation enables dense correspondence estimation across views and time, and compelling video editing applications such as motion exaggeration.

We present a new mid-level patch-based surface representation. At the level of patches, objects across different categories share similarities, which leads to more generalizable models. We show that our representation trained on one category of objects from ShapeNet can also well represent detailed shapes from any other category. In addition, it can be trained using much fewer shapes, compared to existing approaches. We show several applications of our new representation, including shape interpolation and partial point cloud completion. Due to explicit control over positions, orientations and scales of patches, our representation is also more controllable compared to object-level representations, which enables us to deform encoded shapes non-rigidly.

We propose a general-purpose DEep MEsh Autoencoder (DEMEA) which adds a novel embedded deformation layer to a graph-convolutional mesh autoencoder. We demonstrate multiple applications of DEMEA, including non-rigid 3D reconstruction from depth and shading cues, non-rigid surface tracking, as well as the transfer of deformations over different meshes.

We introduce the first dense neural non-rigid structure from motion (N-NRSfM) approach which can be trained end-to-end in an unsupervised manner from 2D point tracks. We formulate the deformation model by an auto-decoder and impose subspace constraints on the recovered latent space function in the frequency domain, allowing us to recover the period of the input sequence. Our method enables multiple applications including shape compression, completion and interpolation, among others. Combined with an encoder trained directly on 2D images, we perform scenario-specific monocular 3D shape reconstruction at interactive frame rates.

We introduce a supervised-learning framework for non-rigid point set alignment of a new kind - Displacements on Voxels Networks (DispVoxNets) - which abstracts away from the point set representation and regresses 3D displacement fields on regularly sampled proxy 3D voxel grids. Thanks to recently released collections of deformable objects with known intra-state correspondences, DispVoxNets learn a deformation model and further priors (e.g., weak point topology preservation) for different object categories such as cloths, human bodies and faces.

We show that an adversary can be trained to control a deep reinforcement learning agent. Our technique works on fully trained victim agents and makes them pursue an alternative, adversarial goal when under attack. In contrast to traditional attacks on e.g. image classifiers, our setting involves adversarial goals that may not be immediately reachable but instead may require multiple steps to be achieved.

Education

Positions

Invited Talks

Miscellaneous

This website is built on Jon Barron's website template!