Computer Vision & Transdisciplinary AI |

U-Net-Based Prediction of Concealed Volume Loss in AZ31 From Optical Corrosion Imprints and As-Exposed Surface Topography

Fri, 09 Oct 2026 00:00:00 +0000

An efficient screening of dissolution modulators of Mg alloys: Combining high-throughput multi-well exposure with topographical quantification of volume loss and pitting corrosion

Wed, 01 Jan 2025 00:00:00 +0000

Unleashing the power of AI in predicting the technological and phenolic maturity of pomegranates cultivated in Lebanon

Wed, 01 Jan 2025 00:00:00 +0000

AI for predicting the technological and phenolic maturity of pomegranates in Lebanon

Sun, 01 Jan 2023 00:00:00 +0000

Pomegranates are one of Lebanon’s important fruit crops, and the date on which they are harvested decides both their health value — their content of antioxidant polyphenols — and their market value. Farmers usually harvest on experience or by the calendar, without accounting for how strongly the climate and the growing region shape fruit maturity, and the laboratory measurements that would tell them are slow and need equipment most farms do not have.

Bringing together food science at Beirut Arab University (Nada El Darra) and AI at the University of Bonn, this interdisciplinary project of the Arab-German Young Academy of Sciences and Humanities (AGYA) asked whether simple, cheap measurements can predict the quality of the fruit before it is harvested.

What we did

A unique field dataset. Over three months of the 2023 season (September–November), 548 sweet pomegranates were harvested from three regions of Lebanon — Hasbaya, El Jahliye and Rachiine. Each fruit was photographed with a smartphone under controlled conditions and weighed and measured, and its juice was analysed in the laboratory for total soluble solids, titratable acidity, maturity index, polyphenol content and colour intensity. Climate data (temperature, solar radiation, humidity, precipitation) were added for each site, with support from the Lebanese Agricultural Research Institute (LARI).
Predicting quality from simple features. Linear regression and multi-layer perceptron models predicted the food-quality indicators from physical, geographical and environmental features. The MLP reached an R² of 0.84 for titratable acidity, 0.70 for phenolic content and 0.65 for colour intensity, and classified fruit into low, moderate and high polyphenol levels with 71% accuracy.
Region matters. Principal component analysis showed clear geographic differences: El Jahliye fruit had low to moderate polyphenol levels, Rachiine moderate to high, and Hasbaya the full range — so harvest strategies should be region-specific. Comparing technological, phenolic and actual harvest dates gave recommended harvest windows per region.

The results were published in Scientific Reports ( ), the first study in Lebanon to use machine learning to predict food-quality indicators of pomegranates.

Perspective

The smartphone images collected in the project open the way to purely image-based maturity assessment in the field — an affordable tool that farmers in Lebanon and similar agro-ecological regions could use to decide when to harvest. The work was also partially supported by the DAAD with funds from the Federal Foreign Office.

Determining a pose of an object in the surroundings of the object by means of multi-task learning

Tue, 01 Feb 2022 00:00:00 +0000

Eyesight to AI: Discovery of efficient corrosion modulators via predictive machine learning models

Fri, 01 Jan 2021 00:00:00 +0000

The fundamental concept of the project is to develop a pattern recognition routine that enables high-throughput quantification of the effect of small organic additives on the degradation of a magnesium alloy by automated classification of corrosion imprints. Subsequently, the quantified optical signal will be used as a target parameter for different supervised and semi-supervised learning approaches to predict the performance of untested additives. Accuracy and robustness of the developed models will be validated by experimental blind tests.

6D Camera Relocalization in Ambiguous Scenes via Continuous Multimodal Inference

Sat, 01 Aug 2020 00:00:00 +0000

Uncertainty Aware Methods for Camera Pose Estimation and Relocalization

Wed, 01 Jan 2020 00:00:00 +0000

Camera pose estimation is the term for determining the 6-DoF rotation and translation parameters of a camera. It is now a key technology in enabling multitudes of applications such as augmented reality, autonomous driving, human computer interaction and robot guidance. For decades, vision scholars have worked on finding the unique solution of this problem. Yet, this trend is witnessing a fundamental change. The recent school of thought has begun to admit that for our highly complex and ambiguous real environments, obtaining a single solution is not sufficient. This has led to a paradigm shift towards estimating rather a range of solutions in the form of full probability or at least explaining the uncertainty of camera pose estimates. Thanks to the advances in Artificial Intelligence, this important problem can now be tackled via machine learning algorithms that can discover rich and powerful representations for the data at hand. In collaboration, TU Munich and Stanford University plan to devise and implement generative methods that can explain uncertainty and ambiguity in pose predictions. In particular, our aim is to bridge the gap between 6DoF pose estimation either from 2D images/3D point sets and uncertainty quantification through multimodal variational deep methods.

Method for determining a pose of an object in an environment of the object using multi task learning and control device

Mon, 01 Jul 2019 00:00:00 +0000

Adversarial Networks for Camera Pose Regression and Refinement

Tue, 01 Jan 2019 00:00:00 +0000

Determination of the pose of an x-ray unit relative to an object using a digital model of the object

Mon, 01 Oct 2018 00:00:00 +0000

Scene coordinate and correspondence learning for image-based localization

Mon, 01 Jan 2018 00:00:00 +0000

When regression meets manifold learning for object recognition and pose estimation

Mon, 01 Jan 2018 00:00:00 +0000

CAD model-based detection and tracking of large objects

Sun, 01 Jan 2017 00:00:00 +0000

Inspecting, assembling and servicing large industrial objects — gas turbines, generators, machine parts — requires knowing precisely where the object is and from where it is observed. In industry, nominal CAD models of these objects are almost always available, yet classical pipelines rely on pre-calibrated set-ups, hand-crafted features and exhaustive search, which break down for mobile devices, textureless metal surfaces, symmetric parts and partial views. The project asked how deep learning can use the CAD model as prior knowledge to estimate object and camera poses directly from images, so that 3D reconstruction, inspection and augmented-reality guidance become possible without manual calibration.

What we did

X-ray PoseNet for mobile X-ray devices. For industrial computed tomography with uncalibrated mobile X-ray systems, a network learns the mapping from X-ray projections — simulated from the object’s CAD model — to the 6-DoF pose of the device, with an additional reconstruction loss that improves the quality of the resulting 3D volume and generalises to unseen instances of the same object class ( ); the approach was also filed as a patent with Siemens ( ).
Object recognition and pose from depth images. A multi-task framework combines manifold learning of viewpoint descriptors (triplet loss) with direct pose regression, so poses can be read out directly or by descriptor matching; the learned descriptors were far more discriminative, with almost 30% higher relative pose accuracy than related work ( ; patents and ).
Camera relocalisation from a single image. Pixel-wise scene-coordinate regression with learned confidences discards erroneous correspondences before pose estimation and supports hypothesis scoring and refinement ( ), and an adversarial framework learns the joint distribution of images and camera poses so that a discriminator can refine the regressed pose ( ).

Our role

Scientific lead of the project as postdoc at TUM CAMP under Nassir Navab (PI), with Slobodan Ilic at Siemens Corporate Technology, and mentoring of Mai Bui’s PhD, which grew out of this collaboration.

Perspective

The work on ambiguity in pose estimation continued in the with Stanford University, which moved from single pose estimates to full distributions over camera poses (ECCV 2020).

X-Ray PoseNet: 6 DoF pose estimation for mobile X-Ray devices

Sun, 01 Jan 2017 00:00:00 +0000

Gradient projection for regularized cryo-electron tomographic reconstruction

Thu, 01 Jan 2015 00:00:00 +0000