The Computer Vision Group headed by Prof. Michael Moeller conducts research in the field of mathematical image processing, computer vision, and machine learning.
Research in computer vision, machine learning, optimization.
Robustness and exploration of variational and machine learning approaches to inverse problems: An overview
Providing an overview of current approaches for solving inverse problems in imaging using variational methods and machine learning.
3D Shape Completion with Test-Time Training
This work addresses the problem of shape completion while using test-time-training during inference.
Evaluation of Video-Assisted Annotation of Human IMU Data Across Expertise, Datasets, and Tools
Comprehensive study on the quality of annotations provided by expert versus novice annotators for inertial-based activity benchmark datasets.
Convergent Data-Driven Regularizations for CT Reconstruction
Proving the convergence with regards to the noise level of two regularization methods for 2D parallel beam CT-reconstruction, and investigating the effect of discretization errors at different resolutions.
WEAR: An Outdoor Sports Dataset for Wearable and Egocentric Activity Recognition
Outdoor sports benchmark dataset for both vision- and inertial-based human activity recognition.
Temporal Action Localization for Inertial-based Human Activity Recognition
Demonstrates the applicability of state-of-the-art TAL models for both offline and near-online Human Activity Recognition using raw inertial data as well as pre-extracted latent features as input.
Weak-Annotation of HAR Datasets using Vision Foundation Models
Proposes a novel, clustering-based annotation pipeline based on vision foundation models to significantly reduce the amount of data that needs to be annotated by a human annotator during creation of human activity recongition datasets.
Implicit Representations for Constrained Image Segmentation
Proposing a variety of provable geometric constraints to improve Image Segmentation performance in data scarcity domains.
Surprisingly Strong Performance Prediction with Neural Graph Features
We show how to do interpretable performance prediction that outperforms complex SOTA predictors like graph neural networks using simple and fast graph features.
Text-guided Explorable Image Super-resolution
Using text inputs to disambiguate solutions of image super-resolution..
Vignetting Effects: a Tool to Characterize a Fourier Ptychographic Microscope
Using the vignetting effect to characterize a Fourier Ptychographic Microscope.
Hang-Time HAR: A Benchmark Dataset for Basketball Activity Recognition Using Wrist-Worn Inertial Sensors
Basketball activity recognition dataset for benchmarking inertial-based human activity recongition approaches.
CCuantuMM: Cycle-Consistent Quantum-Hybrid Matching of Multiple Shapes
We utilize quantum annealing to solve optimization problems in jointly matching multiple, non-rigidly deformed 3D shapes.
QuAnt: Quantum Annealing with Learnt Couplings
In this paper we propose to learn QUBO forms for quantum annealing from data through gradient backpropagation instead of deriving them. As a result, the solution encodings can be chosen flexibly and compactly.
A Data-Driven Study on the Hawthorne Effect in Sensor-Based Human Activity Recognition
Data-driven approach to measure the effects of observation on participants’ execution of five workout-based activities.
An Evaluation of Zero-Cost Proxies - from Neural Architecture Performance to Model Robustness
We analyze the ability of common zero-cost proxies to serve as performance predictors for robustness in a popular NAS search space.
Differentiable Architecture Search: a One-Shot Method?
We investigate differentiable architecture search for the design of novel architectures for inverse problems in a systematic case study.
Evaluating Adversarial Robustness of Low dose CT Recovery
Adversarial attacks on CT recovery networks can still maintain measurement consistency, and could be used to generate diagnostically different solutions.
Exploring Open Domain Image Super-Resolution through Text
Exploring solutions of image super-resolution using pretrained text-to-image diffusion models.
Implicit Representations for Image Segmentation
Proposing convexity constraint to improve Image Segmentation performance in data scarcity domains.
Improving Native CNN Robustness with Filter Frequency Regularization
We study the frequencies in learned convolution filters and achieve improved native robustness with frequency regularization in learned convolution weights.
Kissing to Find a Match: Efficient Low-Rank Permutation Representation
We propose to tackle the curse of dimensionality of large permutation matrices by approximating them using low-rank matrix factorization, followed by a nonlinearity. To this end, we rely on the Kissing number theory to infer the minimal rank required for representing a permutation matrix of a given size, which is significantly smaller than the problem size.
Learning Posterior Distributions in Underdetermined Inverse Problems
An unpaired learning approach for learning posterior distributions of underdetermined inverse problems using two normalizing flows.
On the Direct Alignment of Latent Spaces
We show that aligning the latent space of pretrained models with a linear transformation.
On the Unreasonable Vulnerability of Transformers for Image Restoration - and an easy fix
We show Transformer based restoration networks are not robust, and uncover effects of different attention mechanisms and nonlinearities on adversarially robust generalization.
Sigma: Scale-invariant global sparse shape matching
We propose a novel mixed-integer programming (MIP) formulation for generating precise sparse correspondences for highly non-rigid shapes.
Subsurface Defect Detection and Classification in 3D THz Images of Glass Fiber Reinforced Thermoplastic Based on 3D Convolutional Neural Network
a 3D convolutional neural network (3D CNN) to classify subsurface defects in a glass fiber reinforced thermoplastic (GFRT) composite material inspected by a 3D THz imaging system.
Investigating (re)current state-of-the-art in human activity recognition datasets
Investigates the necessity of recurrent layers - more specifically Long Short-Term Memory layers - in common Deep Learning architectures for Human Activity Recognition.
A Public Repository to Improve Replicability and Collaboration in Deep Learning for HAR
Investigates the necessity of recurrent layers - more specifically Long Short-Term Memory layers - in common Deep Learning architectures for Human Activity Recognition.
Quantum Motion Segmentation
We introduce the first algorithm for motion segmentation that uses quantum annealing.
A generative model for generic light field reconstruction
We train a generative autoencoder for light fields and use it as a prior for a variety of light field reconstruction tasks.
A Simple Strategy to Provable Invariance via Orbit Mapping
We make neural networks invariant by modifying the input pose such that every element from the orbit of transformations maps to the same canonical element..
LDEdit: Towards generalized text guided image manipulation via latent diffusion models
Use Latent Diffusion Models for zero-shot text guided manipulation using DDIM sampling.
On Adversarial Robustness of Deep Image Deblurring
Imperceptible distortion can significantly degrade the performance of SOTA deblurring networks, even producing drastically different content in the output.
Physical Representation Learning and Parameter Identification from Video Using Differentiable Physics
We investigate the combination of differentiable physics and spatial transformers in a deep action conditional video representation network
Stochastic Training is Not Necessary for Generalization
Models trained with full-batch gradient descent and explicit regularization can match the generalization performance of models trained with stochastic minibatching.
Tutorial on Deep Learning for Human Activity Recognition
Full-day Tutorial held at the 2021 ACM International Symposium on Wearable Computers and International Joint Conference on Pervasive and Ubiquitous Computing.
Improving Deep Learning for HAR with Shallow LSTMs
(🏆 Best Paper Award)Proposes Shallow DeepConvLSTM architecture, employing a 1-layered instead of a 2-layered LSTM, which significantly improves results.
Q-Match: Iterative Shape Matching via Quantum Annealing
We develop an iterative method to tackle quadratic assignment problems with quantum annealing. Using this we solve quadratic assignment problems from shape matching.
Witches’ Brew: Industrial Scale Data Poisoning via Gradient Matching
Data poisoning attacks that successfully poison neural networks trained from scratch, even on large-scale datasets like ImageNet.
Adiabatic Quantum Graph Matching with Permutation Matrix Constraints
We develop various methods to tackle graph matching problems with quantum annealing. Using this we solve quadratic assignment problems from shape matching.
Truth or backpropaganda? An empirical investigation of deep learning theory
We call into question commonly held beliefs regarding the loss landscape, optimization, network width, and rank.
Parametric Majorization for Data-Driven Energy Minimization Methods
A new strategy to optimize the bi-level problems arising in training parameterized energy minimization models.
Lifting Layers: Analysis and Applications
We propose a novel non-linear transfer function called lifting, perform theoretical analysis of lifting layer and demonstrate its effectiveness in deep learning approaches to image classification and denoising.
Learning proximal operators: Using denoising networks for regularizing inverse imaging problems
We replace the proximal operator of the regularization used in many convex energy minimization algorithms by a denoising neural network which serves as an implicit natural image prior.
Collaborative Total Variation: A General Framework for Vectorial TV Models
We propose a novel class of regularizations collaborative total variation (CTV), provide theoretical characterization, demonstrate practical application in inverse problems.
Fast sparse reconstruction: Greedy inverse scale space flows
We propose a new greedy sparse recovery method, which approximates L1 minimization more closely
Nonlinear spectral analysis via one-homogeneous functionals: overview and future prospects
We present the motivation and theory of nonlinear spectral representations, based on convex regularizing functionals.
On the implementation of collaborative TV regularization: Application to cartoon+ texture decomposition
Analysis, implementation, and comparison of several vector-valued total variation (TV) methods that extend the Rudin-Osher-Fatemi variational model to color images.
Spectral decompositions using one-homogeneous functionals
We discuss the use of absolutely one-homogeneous regularization functionals in a variational, scale space, and inverse scale space setting to define a nonlinear spectral decomposition of input data
Sublabel-Accurate Convex Relaxation of Vectorial Multilabel Energies
We propose the first sub-label accurate convex relaxation for vectorial multilabel problems by approximating the dataterm in a piecewise convex (rather than piecewise linear) manner.
Sublabel-accurate relaxation of nonconvex energies
We propose a novel spatially continuous framework for convex relaxations based on functional lifting which can be interpreted as a sublabel–accurate solution to multilabel problems.