{"id":700,"date":"2022-05-06T12:29:03","date_gmt":"2022-05-06T11:29:03","guid":{"rendered":"http:\/\/lovc.cs.uni-bonn.de\/?page_id=700"},"modified":"2026-07-17T08:51:05","modified_gmt":"2026-07-17T07:51:05","slug":"research","status":"publish","type":"page","link":"https:\/\/lovc.cs.uni-bonn.de\/index.php\/research\/","title":{"rendered":"Research"},"content":{"rendered":"<style>\n.research_container {\n    display: flex;\n    flex-wrap: wrap;\n    justify-content: left;\n}\n.research-header-wrapper {\n  margin-bottom: 10px;\n}\n.research_div {\n   width: 300px;\n   text-align: justify;\n   margin-left: 0px;\n   margin-right: 30px;\n   margin-top: 5px;\n   margin-bottom: 25px;\n}\n.research_header {\n  color: #000000;\n  font-weight: bold;\n  border-top: 1px solid; \n  border-bottom: 1px solid; \n  border-color: #b5bdbc;\n  width: 300px;\n  text-align: center;\n  font-size: 14px;\n  height: 35px;\n  vertical-align: middle;\n  display: table-cell;\n}\n.research_div figure {\n    margin: 0;\n}\n.research_div img {\n    display: block;\n    border:none;\n    border-style: none;\n    width: 300px;\n    height: 120px;      \/* same height for every image *\/\n    object-fit: contain; \/* preserves aspect ratio *\/\n    margin: 0 auto;\n}\n.research_div figcaption {\n    margin-top: 0px;\n}\n<\/style>\n<p>Below we summarise our high-level interests together with specific research directions.<\/p>\n<div class=\"research_container\">\n<div class=\"research_div\">\n<div class=\"research-header-wrapper\">\n<div class=\"research_header\">Artificial Intelligence &#038; Machine Learning<\/div>\n<\/div>\n<figure>\n<img decoding=\"async\" src=\"\nhttps:\/\/lovc.cs.uni-bonn.de\/\/wp-content\/uploads\/2026\/07\/chi.png\" alt=\"\"\/><figcaption>\nWe are interested in the <strong>foundations<\/strong> of AI and ML, and in solving specific tasks by <strong>inferring general principles from data<\/strong>.<\/p>\n<details>\n<summary>\nMore<br \/>\n<\/summary>\n<p>For example, we develop methods for <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2405.17897\">multi-model merging<\/a>, <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2601.03420\">LLM jailbreaking<\/a>, <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2307.13078\">certified training<\/a>, or <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2205.06688.pdf\">differentiable programming<\/a>.<br \/>\n<\/figcaption><\/figure>\n<\/div>\n<div class=\"research_div\">\n<div class=\"research-header-wrapper\">\n<div class=\"research_header\">Computer Vision &#038; Image Understanding<\/div>\n<\/div>\n<figure>\n<img decoding=\"async\" src=\"\nhttps:\/\/lovc.cs.uni-bonn.de\/\/wp-content\/uploads\/2026\/07\/pix2lr-1.png\" alt=\"\"\/><figcaption>\nWe are excited about leveraging fundamental principles (e.g. about 3D geometry or physics) to push the boundaries of <strong>image and video understanding<\/strong>.<\/p>\n<details>\n<summary>\nMore<br \/>\n<\/summary>\n<p>Examples include <a href=\"https:\/\/arxiv.org\/pdf\/2607.05006\" target=\"_blank\">pixel-wise left-right image understanding<\/a>, <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2504.12825\">image-driven physically plausible 4D generation<\/a>, <a target=\"_blank\"  href=\"https:\/\/openaccess.thecvf.com\/content\/CVPR2024\/papers\/Wang_Unsupervised_3D_Structure_Inference_from_Category-Specific_Image_Collections_CVPR_2024_paper.pdf\">unsupervised 3D structure inference in images<\/a>, or <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2204.14030\">physics inference from video<\/a>.<\/p>\n<\/details>\n<\/figcaption><\/figure>\n<\/div>\n<div class=\"research_div\">\n<div class=\"research-header-wrapper\">\n<div class=\"research_header\">Geometry &#038; Shape Analysis<\/div>\n<\/div>\n<figure>\n<img decoding=\"async\" src=\"https:\/\/lovc.cs.uni-bonn.de\/\/wp-content\/uploads\/2025\/11\/sang3dv2026.png\" alt=\"\"\/><figcaption>\nWe develop methods to analyse and model <strong>geometric data<\/strong>, e.g. graphs, 3D shapes, 4D sequences, and high-dimensional manifolds. <\/p>\n<details>\n<summary>\nMore<br \/>\n<\/summary>\n<p>Examples include <a target=\"_blank\"  href=https:\/\/arxiv.org\/pdf\/2509.04145>generative models based on hyper networks and diffusion<\/a>, <a target=\"_blank\"  href=https:\/\/www.arxiv.org\/pdf\/2508.05505>self-supervised symmetry understanding<\/a>, <a target=\"_blank\"  href=https:\/\/openaccess.thecvf.com\/content\/CVPR2026\/papers\/Ehm_Teaching_DINOv3_About_Partial_3D_Geometry_A_Self-Supervised_Geometry-Aware_Approach_CVPR_2026_paper.pdf>self-supervised foundation model fine-tuning for partial 3D shapes<\/a>, <a target=\"_blank\"  href=https:\/\/arxiv.org\/pdf\/2411.03511> 3D shape matching benchmarks<\/a>, <a target=\"_blank\"  href=https:\/\/arxiv.org\/pdf\/2504.06385>globally optimal and geometrically consistent 3D shape matching<\/a>, and <a target=\"_blank\"  href=https:\/\/nafieamrani.github.io\/assets\/pdf\/nafie2024miccai.pdf> statistical shape models<\/a>.<br \/>\n<\/details>\n<\/figcaption><\/figure>\n<\/div>\n<div class=\"research_div\">\n<div class=\"research-header-wrapper\">\n<div class=\"research_header\">Mathematical Modelling &#038; Optimisation<\/div>\n<\/div>\n<figure>\n<img decoding=\"async\" src=\"\nhttps:\/\/lovc.cs.uni-bonn.de\/\/wp-content\/uploads\/2026\/07\/fastmrf.png\" alt=\"\"\/><figcaption>\nWe are excited about mathematical and algorithmic foundations, in many cases seeking for <strong>theoretical guarantees<\/strong>.<\/p>\n<details>\n<summary>\nMore<br \/>\n<\/summary>\n<p>Theoretical guarantees include global optimality, cycle consistency or neural network robustness.<br \/>\nFor example, we develop globally optimal methods for <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2504.06385\">3D shape matching<\/a>, <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2211.11589\">2D-3D matching<\/a>, or <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2110.00053.pdf\">sparse optimisation over the Stiefel manifold<\/a>. We develop methods that achieve cycle consistency for <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2212.00780\">multi-graph matching<\/a> and <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/1410.8546.pdf\">multi-alignment<\/a>, for example through <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/1811.10541.pdf\">higher-order projected power iterations<\/a>, <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/1803.06320.pdf\">non-negative matrix factorisation<\/a>, or <a target=\"_blank\"  href=\"http:\/\/openaccess.thecvf.com\/content_CVPR_2019\/papers\/Swoboda_A_Convex_Relaxation_for_Multi-Graph_Matching_CVPR_2019_paper.pdf\">convex relaxations<\/a>. We also tackle <a target=\"_blank\"  href=https:\/\/arxiv.org\/pdf\/2307.13078>certified neural network training<\/a>.<br \/>\n<\/details>\n<\/figcaption><\/figure>\n<\/div>\n<div class=\"research_div\">\n<div class=\"research-header-wrapper\">\n<div class=\"research_header\">Geometry Optimisation<\/div>\n<\/div>\n<figure>\n<img decoding=\"async\" src=\"https:\/\/lovc.cs.uni-bonn.de\/\/wp-content\/uploads\/2026\/07\/lb1_boundary_web.gif\" alt=\"\"\/><figcaption>\nWe are interested in <strong>discovering novel geometries<\/strong> that have specific physical properties.<\/p>\n<details>\n<summary>\nMore<br \/>\n<\/summary>\n<p>For example, we are working on designing stellerator geometries for nuclear fusion based on physics simulation. Also, we are interested in designing mechanical meta-materials through geometry optimisation.<br \/>\n<\/details>\n<\/figcaption><\/figure>\n<\/div>\n<div class=\"research_div\">\n<div class=\"research-header-wrapper\">\n<div class=\"research_header\">Physics-Grounded Visual Computing<\/div>\n<\/div>\n<figure>\n<img decoding=\"async\" src=\"\nhttps:\/\/lovc.cs.uni-bonn.de\/\/wp-content\/uploads\/2026\/07\/physicsgroundedvisualcomputing.drawio.jpg\" alt=\"\"\/><figcaption>\nWe <strong>exploit physics as a prior<\/strong> to tackle diverse visual computing tasks. Also, we <strong>infer physics<\/strong> from visual data.<\/p>\n<details>\n<summary>\nMore<br \/>\n<\/summary>\n<p>Examples include<br \/>\n<a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2504.12825\">image-driven physically plausible 4D generation<\/a>, or  <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2204.14030\">physics inference and physics-based video editing<\/a>.<br \/>\n<\/details>\n<\/figcaption><\/figure>\n<\/div>\n<div class=\"research_div\">\n<div class=\"research-header-wrapper\">\n<div class=\"research_header\">Correspondence Problems<\/div>\n<\/div>\n<figure>\n<img decoding=\"async\" src=\"https:\/\/lovc.cs.uni-bonn.de\/\/wp-content\/uploads\/2026\/06\/cao2026eccv-10.png\" alt=\"\"\/><figcaption>\nWe develop <strong>optimisation and machine learning<\/strong> approaches for identifying corresponding parts across objects.<\/p>\n<details>\n<summary>\nMore<br \/>\n<\/summary>\n<p>We work on <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2103.17229\">image correspondences<\/a>, <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2207.00291.pdf\">graph matching<\/a>, <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2212.00780\">multi-graph matching<\/a>, <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2407.08244\">3D shape matching<\/a>, <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2207.09610\">multi-shape matching<\/a>, etc.<br \/>\n<\/details>\n<\/figcaption><\/figure>\n<\/div>\n<div class=\"research_div\">\n<div class=\"research-header-wrapper\">\n<div class=\"research_header\">Graph-Based Modelling &#038; Graph Algorithms<\/div>\n<\/div>\n<figure>\n<img src=https:\/\/lovc.cs.uni-bonn.de\/\/wp-content\/uploads\/2026\/07\/gecograph.png alt=\"\"\/><figcaption>\nMany visual computing problems can be formalised as <strong>graph-theoretic problems<\/strong>, so that they can be solved using <strong>graph algorithms<\/strong>. <\/p>\n<details>\n<summary>\nMore<br \/>\n<\/summary>\n<p>For example, numerous variants of matching problems can be addressed by finding shortest paths in product graphs, e.g. <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2211.11589\">2D-to-3D shape matching<\/a>, <a target=\"_blank\"  href=\"https:\/\/openaccess.thecvf.com\/content\/CVPR2025\/papers\/Roetzer_Higher-Order_Ratio_Cycles_for_Fast_and_Globally_Optimal_Shape_Matching_CVPR_2025_paper.pdf\">graph-to-image matching<\/a>, or <a target=\"_blank\"  href=\"https:\/\/openaccess.thecvf.com\/content\/CVPR2024\/papers\/Roetzer_SpiderMatch_3D_Shape_Matching_with_Global_Optimality_and_Geometric_Consistency_CVPR_2024_paper.pdf\">3D shape matching<\/a>.<br \/>\n<\/details>\n<\/figcaption><\/figure>\n<\/div>\n<div class=\"research_div\">\n<div class=\"research-header-wrapper\">\n<div class=\"research_header\">Scalable Algorithms<\/div>\n<\/div>\n<figure>\n<img decoding=\"async\" src=\"https:\/\/lovc.cs.uni-bonn.de\/\/wp-content\/uploads\/2026\/07\/gecotime.png\" alt=\"\"\/><figcaption>\nWe develop <strong>efficient algorithms<\/strong> that scale to very large data collections, which is often a prerequisite for training large-scale AI models.<\/p>\n<details>\n<summary>\nMore<br \/>\n<\/summary>\n<p>For example, we develop scalable algorithms for <a target=\"_blank\"  href=\"https:\/\/openaccess.thecvf.com\/content\/CVPR2026\/papers\/Roetzer_Fast_Markov_Random_Field_Optimisation_for_Topologically_Noisy_3D_Shape_CVPR_2026_paper.pdf\">3D shape matching<\/a>, <a target=\"_blank\"  href=\"https:\/\/openaccess.thecvf.com\/content\/CVPR2025\/papers\/Kahl_Towards_Optimizing_Large-Scale_Multi-Graph_Matching_in_Bioimaging_CVPR_2025_paper.pdf\">multi-graph matching<\/a>, <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/1811.10541.pdf\">multi-shape matching<\/a>, <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2110.00053.pdf\">matrix synchronisation<\/a>, or <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2206.09596\">sub-label accurate energy minimisation<\/a>.<\/p>\n<\/details>\n<\/figcaption><\/figure>\n<\/div>\n<div class=\"research_div\">\n<div class=\"research-header-wrapper\">\n<div class=\"research_header\">Applications<\/div>\n<\/div>\n<figure>\n<img src=https:\/\/lovc.cs.uni-bonn.de\/wp-content\/uploads\/2023\/04\/medical_imaging.jpg alt=\"\"\/><figcaption>\nMany problems that we study are driven by <strong>real-world applications<\/strong>.<\/p>\n<details>\n<summary>\nMore<br \/>\n<\/summary>\n<p>Examples of applications include <a target=\"_blank\"  href=\"https:\/\/nafieamrani.github.io\/assets\/pdf\/nafie2024miccai.pdf\">medical shape models<\/a>, <a target=\"_blank\"  href=\"https:\/\/openaccess.thecvf.com\/content\/CVPR2025\/papers\/Kahl_Towards_Optimizing_Large-Scale_Multi-Graph_Matching_in_Bioimaging_CVPR_2025_paper.pdf\">bioimaging<\/a>, <a target=\"_blank\"  href=\"https:\/\/handtracker.mpi-inf.mpg.de\/projects\/RGB2Hands\/content\/RGB2Hands_author_version.pdf\">3D reconstruction and real-time tracking of interacting hands<\/a>, <a target=\"_blank\"  href=\"https:\/\/arxiv.org\/pdf\/2011.14143.pdf\">morphable head models<\/a>, and <a target=\"_blank\"  href=\"http:\/\/gvv.mpi-inf.mpg.de\/projects\/StyleRig\/data\/paper.pdf\">portrait image editing<\/a>.<br \/>\n<\/details>\n<\/figcaption><\/figure>\n<\/div>\n<\/div>\n","protected":false},"excerpt":{"rendered":"<p>Below we summarise our high-level interests together with specific research directions. Artificial Intelligence &#038; Machine Learning We are interested in the foundations of AI and ML, and in solving specific tasks by inferring general principles from data. More For example, we develop methods for multi-model merging, LLM jailbreaking, certified training, or differentiable programming. Computer Vision [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"parent":0,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"page-no-title","meta":{"footnotes":""},"class_list":["post-700","page","type-page","status-publish","hentry"],"_links":{"self":[{"href":"https:\/\/lovc.cs.uni-bonn.de\/index.php\/wp-json\/wp\/v2\/pages\/700","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/lovc.cs.uni-bonn.de\/index.php\/wp-json\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/lovc.cs.uni-bonn.de\/index.php\/wp-json\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/lovc.cs.uni-bonn.de\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/lovc.cs.uni-bonn.de\/index.php\/wp-json\/wp\/v2\/comments?post=700"}],"version-history":[{"count":231,"href":"https:\/\/lovc.cs.uni-bonn.de\/index.php\/wp-json\/wp\/v2\/pages\/700\/revisions"}],"predecessor-version":[{"id":2639,"href":"https:\/\/lovc.cs.uni-bonn.de\/index.php\/wp-json\/wp\/v2\/pages\/700\/revisions\/2639"}],"wp:attachment":[{"href":"https:\/\/lovc.cs.uni-bonn.de\/index.php\/wp-json\/wp\/v2\/media?parent=700"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}