“Signal Processing in the AI era” was the tagline of this year’s IEEE International Conference on Acoustics, Speech and Signal Processing, taking place in Rhodes, Greece.

In this context, Brent de Weerdt, Xiangyu Yang, Boris Joukovsky, Alex Stergiou and Nikos Deligiannis presented ETRO’s research during poster sessions and oral presentations, with novel ways to process and understand graph, video, and audio data. Nikos Deligiannis chaired a session on Graph Deep Learning, attended the IEEE T-IP Editorial Board Meeting, and had the opportunity to meet with collaborators from the VUB-Duke-Ugent-UCL joint lab.

Featured articles:

• De Weerdt, B. et al., “Designing Transformer Networks for Sparse Recovery of Sequential Data Using Deep Unfolding” (https://ieeexplore.ieee.org/abstract/document/10094712)
• Yang X. et al., “Relevance Propagation through Deep Conditional Random Fields” (https://ieeexplore.ieee.org/abstract/document/10095075)
• Stergiou A. et al., “Play it back: Iterative attention for audio recognition” (https://ieeexplore.ieee.org/abstract/document/10096532, developed at the University of Bristol)

The paper entitled” Novel Multi-Parametric Sensor System for Comprehensive Multi-Wavelength Photoplethysmography Characterization”, published in the journal MDPI SENSORS in 2023, coauthored by: J.L. Cause, Á. Solé-Morillo, B. Da Silva, J. García-Naranjo, and J. Stiens was awarded with the first prize of the provincial council of scientific health societies of Santiago de Cuba

Johan Stiens gave an interview with deputy of the province Flemish-Brabant on Ring TV about the TRN radar in support for renovation in the construction industry

Subsidies van 600.000 euro voor innoverende bedrijven in Vlaams-Brabant | Ring (ringtv.be)

600.000 euro voor innoverende Vlaams-Brabantse bedrijven | Voka

On June 27th 2024 at 09:30, Cheng Chen will defend their PhD entitled “NOVEL FABRICATION AND ELECTROMAGNETIC-OPTICAL CHARACTERIZATION TECHNIQUES OF CARBON-BASED NANOSTRUCTURES”.

Everybody is invited to attend the presentation in room D.2.01, or digitally via this link.

Nowadays, carbon nanomaterials are increasingly garnering attention as the next generation of semiconductor materials. Notably, graphene and carbon nanofibers (CNFs) have emerged as pivotal players in the semiconductor domain, owing to their remarkable electrical, mechanical, and thermal properties, coupled with their distinctive structural attributes.

Graphene, characterized by its two-dimensional single-layer structure of densely packed carbon atoms, boasts unparalleled electrical conductivity. This positions it for significant potential in applications like highfrequency electronic devices and sensors. Furthermore, its transparency and flexibility pave the way for innovative advancements in flexible electronic devices and display technologies, rejuvenating the electronics industry’s potential. CNFs, celebrated for their nanoscale diameter and exceptional mechanical attributes, carve a niche for themselves in material science. Their superior conductivity heralds vast opportunities, especially in realms such as conductive fibers and flexible circuitry. Within the spectrum of synthesis techniques, Chemical Vapor Deposition (CVD) emerges as a standout method, particularly for producing highquality graphene films and CNFs.

This dissertation delves into the CVD preparation, performance characterization, and subsequent applications of these materials, particularly in electromagnetic (EM) and ultraviolet (UV) optics. Specifically, the research encompasses:

1. Investigating the photoluminescence (PL) performance of ZnO/graphene composite films and ZnO@CNFs composite fibers within the UV emission spectrum
2. Determining the optimal thickness of ZnO in the ZnO/graphene composite for optimal UV-PL behavior
3. Analyzing the anisotropic EM loss behavior of CNFs and ZnO@CNFs samples across the MMW and sub-terahertz bands
4. Pioneering a novel quality-assessment method for wafer-scale CVDprepared graphene films, grounded in their distinctive interactions with terahertz TE10 waves

In essence, this research centers on graphene and CNFs, exploring their potential in the realm of EM and UV optics and offering insights based on their intrinsic properties.

On June 20th 2024 at 16:00, Yangxintong Lyu will defend their PhD entitled “DEEP-LEARNING-BASED MULTI-MODAL FUSION FOR TRAFFIC IMAGE DATA PROCESSING”.

Everybody is invited to attend the presentation in room I.0.02, or digitally via this link.

In recent years, deep-learning-based technologies have significantly developed, which is driven by a large amount of data associated with task-specific labels. Among the various formats used for representing object attributes in computer vision, RGB images stand out as a ubiquitous choice. Their value extends to traffic-related applications, particularly in the realms of autonomous driving and intelligent surveillance systems. By using an autonomous driving system, a car is capable of navigating and operating with diminished human interactions, while traffic conditions can be monitored and analysed by an intelligent system. Essentially, the techniques reduce human error and improve road safety, which significantly impacts our daily life.

Although many visual-based traffic analysis tasks can indeed be effectively solved by leveraging features extracted from a sole RGB channel, certain unresolved challenges persist that introduce extra difficulties under certain situations. First of all, extracting complicated information becomes demanding, especially under erratic lighting conditions, raising a need for auxiliary clues. Secondly, obtaining large-scale accurate labels for challenging tasks remains time-consuming, costly, and arduous. The former prompts exploration into capturing and exploiting additional information such that the objects can be observed from diverse aspects; in contrast, the latter requires either an increase in the volume of available data or the capability to learn from other datasets that already possess perfect labels.

In this thesis, we tackle multi-modal data fusion and data scarcity for intelligent transportation systems. Our first contribution is a novel RGB-Thermal fusion neural network for semantic segmentation. It ensures the segmentation under limited illumination. Our second contribution is a 3D-prior-based framework for monocular vehicle 6D pose estimation. The use of 3D geometry avoids the ill-posed pose prediction from a single camera viewpoint. Thanks to the extra 3D information, our novel method can handle distant and occluded vehicles. The third contribution is a real-world, large-scale vehicle make and model dataset that contains the most popular brands operating in Europe. Moreover, we propose a two-branch deep learning vehicle make and model recognition paradigm to reduce inter-make ambiguity. The last contribution is a weakly supervised vehicle 6D pose estimation paradigm by adapting knowledge built based on a novel synthetic dataset. The dataset includes a large amount of accurate labels for vehicles. By learning from the synthetic dataset, our method allows the significant reduction of expensive real-life vehicle pose annotations.

Comprehensive experimental results reveal that the newly introduced datasets hold significant promise for deep-learning-based processing of traffic image data. Moreover, the proposed methods surpass the existing baselines in the literature. Our research not only yields high-quality scientific publications but also underscores its value across both academic and industrial domains. 

A full immersive experience of Augmented Reality for neurosurgical planning and real-time intervention, demonstrated by Taylor on the FARI immersive CAVE during the Agoria HealthTech roundtable event June 17^th , 2024.

On July 1st 2024 at 16:00, Panagiotis Gonidakis will defend their PhD entitled “DATA- AND LABEL-EFFICIENT DEEP LEARNING FOR MEDICAL IMAGE ANALYSIS APPLICATION TO LUNG NODULE DETECTION ON THORACIC CT”.

Everybody is invited to attend the presentation in room D.0.03, or digitally via this link.

Convolutional neural networks (CNNs) have been widely used to detect and classify various objects and structures in computer vision and medical imaging. Access to large sets of annotated data is commonly a prerequisite for achieving good performance. In medical imaging, acquiring adequate amounts of labelled data can often be time consuming and costly. Therefore, reducing the need for data and in particular associated annotations, is of high importance for medical imaging applications. In this work we investigated whether we can lower the need of annotated data for a supervised learning classification problem.  

We chose to tackle the problem of lung nodule detection in thoracic computed tomography (CT) imaging, as this widely investigated application allowed us to benefit from publicly available data and benchmark our methods. We designed a 3D CNN architecture to perform patch-wise classification of candidate nodules for false positive reduction. Its training, testing and fine-tuning procedure is optimized, we evaluated its performance, and we compared it with other state-of-the-art approaches in the field.  

Next, we explored how data augmentation can contribute towards more accurate and less data-demanding models. We investigated the relative benefit of increasing the amount of original data, with respect to computationally augmenting the amount of training samples. Our result indicated that in general, better performance is achieved when increasing the amount of unique data samples, or augmenting the data more extensively, as expected. Surprisingly however, we observed that after reaching a certain amount of training samples, data augmentation led to significantly better performance compared to adding unique samples. Amongst investigated augmentation methods, rotations were found to be most beneficial for improving model performance. 

Following, we investigated the benefit of combining deep learning with handcrafted features. We explored three fusion strategies with increasing complexity and assessed their performance for varying amounts of training data. Our findings indicated that combining handcrafted features with a 3D CNN approach significantly improved lung nodule detection performance in comparison to an independently trained CNN model, regardless of the fusion strategy. Comparatively larger increases in performance were obtained when less training data was available. The fusion strategy in which features are combined with a CNN using a single end-to-end training scheme performed best overall, allowing to reduce training data by 33% to 43%, while maintaining performance. Among the investigated handcrafted features, those that describe the relative position of the candidate with respect to the lung wall and mediastinum, were found to be of most benefit. 

Finally, we considered the case in which abundant data is available, but annotations are scarce, and investigated several methods to improve label-efficiency and their combined effect. We proposed a framework that utilizes both annotated and unannotated data, can be pretrained via self-supervision, and allows to combine handcrafted features with learned representations. Interestingly, the improvements in performance derived from the proposed learning schemes were found to accumulate, leading to increased label-efficiency when these strategies are combined. We observed a potential to decrease the amount of annotated data up to 68% when compared to traditional supervised training, while maintaining performance.  

Our findings indicate that the investigated methods allow considerable reduction of data and/or annotations while maintaining model performance for lung nodule detection from CT imaging. Future work should investigate whether these results generalize to other domains, such that more applications that face challenges due to a shortage of annotated data may benefit from the potential of deep learning.

On June 13th 2024 at 16:00, Remco Royen will defend their PhD entitled “ADDRESSING LABELLING, COMPLEXITY, LATENCY, AND SCALABILITY IN DEEP LEARNING-BASED PROCESSING OF POINT CLOUDS”.

Everybody is invited to attend the presentation in room I.0.01, or digitally via this link.

The field of 3D technology is attracting considerable academic and industrial interest due to its expanding range of potential applications. Noteworthy domains include, but extend beyond, automotive, gaming, extended and augmented reality, drone inspection, robotics, medical imaging, and 3D modeling and design. An essential aspect for these applications is 3D scene understanding. Point clouds, consisting of a collection of points, play a crucial role in capturing the spatial information of physical environments. This results in a lightweight and precise 3D representation, preserving fine details and enabling efficient integration with real-world data.

In recent years, deep learning has gained widespread use, demonstrating its significance across various domains. Its ability to automatically learn intricate patterns from vast datasets has resulted in a transformative impact, driving advancements in technology, and reshaping the landscape of artificial intelligence applications. The ongoing development of increasingly sophisticated neural network architectures continues to push the boundaries of what is achievable across diverse sectors.

As a result, deep learning has become ubiquitous. However, certain limitations hinder its broad applicability. This thesis delves into four crucial challenges associated with deep learning-based point cloud processing: (i) the precise labeling of extensive datasets, (ii) the model complexity requirements, (iii) the latency introduced during inference, and (iv) the concept of scalability. The initial challenge stems from the necessity for extensive datasets with highly accurate annotations. Particularly in the 3D domain, obtaining such high-quality annotations proves challenging and, consequently, expensive. The second challenged arises from the development of more intricate and memory-intensive, facilitated by advancements in high-power-consuming graphics cards. While these methods achieve higher performance levels, they impose constraints on deployment, particularly for embedded devices. Furthermore, the escalating complexity of these networks is accompanied by an increased inference time, impeding real-time applications. Lastly, deep learning-based solutions lack the concept of scalability which have proven vital in traditional methods.

In this thesis, we tackle these challenges and propose diverse solutions within the deep learning paradigm. The thesis commences with the introduction of a rapid 3D LiDAR simulator, designed to mitigate the labeling problem by learning from perfectly annotated synthetic data. We demonstrate its applications in 3D denoising and semantic segmentation. A second contribution can be found within the domain of point cloud instance segmentation. Through the joint learning of prototypes and coefficients, we present an efficient and rapid method that demands relatively low GPU memory. To further improve our method, we introduce an enhanced block merging algorithm. As a third main contribution, we achieve deep learning-based quality scalability by learning embedded latent representations, demonstrating compelling results in applications such as image reconstruction, point cloud compression, and image semantic hashing. The final contribution introduces resolution-scalable 3D semantic segmentation of point clouds. When applied to resolutionscalable 3D sensors, it enables joint point cloud acquisition and processing.

Our proposed methods consistently outperform established benchmarks across diverse datasets, as demonstrated through comprehensive experimentation. The research findings have been disseminated in various reputable journals and conferences, and have led to a patent submission, highlighting their impact in both academic and industrial contexts.

On June 17th 2024 at 10:00, Ruben De Smet will defend their PhD entitled “RAPID PROTOYPING AND DEPLOYMENT OF PRIVACY-ENHANCING TECHNOLOGIES”.

Everybody is invited to attend the presentation in room D.2.01, or digitally via this link.

Since its inception, the internet has quickly become a public service utility. The combination of its commercial exploitation, and the rather intimate nature of how humans actively use the internet, gives rise to some paradoxical situations. As a citizen of Belgium, I would probably not expect to give my name and phone number to a company in the United States to talk to my brother, 50 km up north. However, for over two billion people, this is their rather paradoxical reality: the company Meta, owning WhatsApp, collects and stores these data for their users. This cherry-picked scenario stands example for a wider trend in the industry.

Cryptographers have worked on several privacy-enhancing technologies (PETs). These PETs aim to minimize the amount of personal data to fulfil a service for users. Although these technologies exist on paper, several practical issues arise. These practicalities are the subject of this thesis.

One practical issue is the performance. PETs that run on end-user devices should both be fast and require little bandwidth. We investigate how implementation details may lead to significant speedups or bandwidth savings. Specifically, we devise a zero-knowledge proof (ZKP) tailored to electronic road pricing (ERP). ERP is a privacy-sensitive topic, and our ERP system achieves some notable performance improvements over preexisting proposals.

A second practical issue is the challenging nature of implementing PETs. We present “Circuitree” and “Whisperfish”, to study how to bring PETs to an actual application. Circuitree is a high-level framework to tailor ZKPs to specific scenarios, using a bespoke logic programming language. The language is designed such that the resulting ZKP is highly efficient.

Whisperfish is effectively a reimplementation of the Signal instant messaging client, and allows us to present in detail how Signal deploys their PETs to users. All ideas put forward in this thesis were evaluated by means of their implementation in the Rust programming language.

… Nvidia is the undisputed leader in GPUs, the processors capable of handling very complex calculations. Nvidia currently has a significant amount of cash on hand, which is useful for making substantial investments and staying ahead….

https://www.tijd.be/ondernemen/technologie/houdt-momentum-aan-voor-nvidia-pas-als-technologie-radicaal-verandert-kan-de-chipreus-in-de-problemen-komen/10547908.html

https://flux50.com/news-events/events/15de-energiecongres

Johan Stiens is a guest speaker at this 15^th Energy Congress, entitled CLIMATE-FIT by innovation iN BUILDING CONSTRUCTION on May 30^th in Mechelen, with the invited talk: INTOWALL- radar technology for inspection of the building envelope. 

On June 6th 2024 at 16:00, Anirudh Praveen Kankuppe Raghavendra Swamy will defend their PhD entitled “LOW POWER MM-WAVE FMCW RADAR RECEIVERS IN V AND D BANDS”.

Everybody is invited to attend the presentation in room D.2.01, or digitally via this link.

From the beginning of 20th Century, RADARs have been the cornerstone of the military arsenal. Radar has served as a prime sensing and tracking device and has evolved in complexity over time, while relying on simple modulation like FMCW (frequency-modulated continuous wave). Earlier radars started with operating frequencies of a few MHz, while modern radars operate at frequencies higher than 30 GHz, termed as mm-wave, to benefit from a large available bandwidth. While retaining their fundamental aspect of range and velocity sensing, today’s radars are found in diverse domestic areas like automotive, indoor sensing, human machine interface and vital signs monitoring. Though range and velocity sensing can be achieved with a single radar transceiver, to sense any angle of arrival, they must be arranged as MIMO (multiple input, multiple output) arrays. As the array size grows, it is paramount to have energy efficient array elements that do not compromise performance. The receiver of such an array should be resistant to leakage from the radar transmitter to the receiver termed as spillover. Otherwise, this spillover can result in saturation of the front-end or baseband circuitry in the receiver. Further, to have a better range resolution, the bandwidth of the front-end should be as high as possible. In this work, such a mm-wave radar receiver is explored with an emphasis on low power consumption, large RF bandwidth, robustness to spillover, and unique narrow-band filter for spillover or nearby large target attenuation. Two receivers were designed in a 28 nm bulk CMOS process, operating at 60 GHz (V-band) and 140 GHz (D- band) with a power consumption of 5.2 mW and 67 mW and a bandwidth of 10.2 GHz and 18.3 GHz, respectively. Core of the innovations aiding these state-of-the-art power consumption numbers are a mixer-first front-end architecture, a source-degenerated high-pass filter, a variable gain band-pass Gm-C filter, a low-power broadband I/Q RF front-end at D-band, and a unique, tunable narrow-band spillover and target attenuation filters. With this record low power consumption, the radars have been demonstrated to detect multi-targets, pedestrian movement, heartbeat and could filter selective targets in the range with a 13 mm range resolution marking a spot among state-of-the-art FMCW radar receivers and setting a benchmark for the future.

On May 30th 2024 at 15:00, Taylor Frantz will defend their PhD entitled “AUGMENTED REALITY IN SURGERY ON THE DEVELOPMENT OF REAL-TIME INTERVENTIONAL PLANNING AND NAVIGATION FOR NEUROSURGICAL AND ORTHOPEDIC USE CASES: BENCH-TOP TO CLINICAL EVALUATION”.

Everybody is invited to attend the presentation in room D.2.01, or digitally via this link.

Computer aided navigation (CAN) is a surgical technology which allows a surgeon to use patient medical image data as map to guide the procedure. It comprises several interconnected processes: visualization of 3D medical image data, tracking of surgical instrumentation, definition of a virtual coordinate system around the patient, and the alignment of the image data to the patient.

Despite quantitative benefits, the technology is often not used due to size, cost, and unintuitive visualization of 3D patient data as 2D black and white image. Augmented reality (AR) devices often integrate requisite hardware for CAN into a compact and mobile head mounted device (HMD) and allow the surgeon to view complex 3D data as a hologram overlying the patient. This work addresses technical limitations of such low-cost AR hardware with respect to tracking performance and presents evidence supporting their use in both neurosurgical and orthopedic domains.

Building on early work in videometric tracking as a proof-of-concept, the development of monocular infrared (IR) tracking of existing surgical instrumentation provided a method to establish a room-stable coordinate system and a mechanism for precise user input; both required for CAN. This tracking solution was validated in a VICON motion capture lab and demonstrated a mean pose estimation error of 0.78 mm ± 0.74 mm and 0.84° ± 0.64°.

Following this, phantom trials in navigated external ventricular drain (EVD) placement, and total shoulder and hip arthroplasty were performed. The results demonstrated a reduction in technique learning curve of the former, and improved outcomes of the latter when compared to traditional non-navigated techniques. Moreover, AR data registration was found to be comparable to modern CAN systems.

Clinical trials in both tumor resection planning and EVD were then performed to assess efficacy of AR-CAN compared to current surgical practice. In the former, AR-CAN demonstrated a reduction in preoperative planning time with superior lesion delineation when compared to neuronavigation. Preliminary results in AR navigated EVD placement outcomes demonstrate 82% optimal (grade I), 18% sub optimal (grade II), and 0% (grade III). This currently outperforms literature, given single attempt insertion.

“Signal Processing in the AI era” was the tagline of this year’s IEEE International Conference on Acoustics, Speech and Signal Processing, taking place in Rhodes, Greece.

In this context, Brent de Weerdt, Xiangyu Yang, Boris Joukovsky, Alex Stergiou and Nikos Deligiannis presented ETRO’s research during poster sessions and oral presentations, with novel ways to process and understand graph, video, and audio data. Nikos Deligiannis chaired a session on Graph Deep Learning, attended the IEEE T-IP Editorial Board Meeting, and had the opportunity to meet with collaborators from the VUB-Duke-Ugent-UCL joint lab.

Featured articles:

• De Weerdt, B. et al., “Designing Transformer Networks for Sparse Recovery of Sequential Data Using Deep Unfolding” (https://ieeexplore.ieee.org/abstract/document/10094712)
• Yang X. et al., “Relevance Propagation through Deep Conditional Random Fields” (https://ieeexplore.ieee.org/abstract/document/10095075)
• Stergiou A. et al., “Play it back: Iterative attention for audio recognition” (https://ieeexplore.ieee.org/abstract/document/10096532, developed at the University of Bristol)

The paper entitled” Novel Multi-Parametric Sensor System for Comprehensive Multi-Wavelength Photoplethysmography Characterization”, published in the journal MDPI SENSORS in 2023, coauthored by: J.L. Cause, Á. Solé-Morillo, B. Da Silva, J. García-Naranjo, and J. Stiens was awarded with the first prize of the provincial council of scientific health societies of Santiago de Cuba

Johan Stiens gave an interview with deputy of the province Flemish-Brabant on Ring TV about the TRN radar in support for renovation in the construction industry

Subsidies van 600.000 euro voor innoverende bedrijven in Vlaams-Brabant | Ring (ringtv.be)

600.000 euro voor innoverende Vlaams-Brabantse bedrijven | Voka

On June 27th 2024 at 09:30, Cheng Chen will defend their PhD entitled “NOVEL FABRICATION AND ELECTROMAGNETIC-OPTICAL CHARACTERIZATION TECHNIQUES OF CARBON-BASED NANOSTRUCTURES”.

Everybody is invited to attend the presentation in room D.2.01, or digitally via this link.

Nowadays, carbon nanomaterials are increasingly garnering attention as the next generation of semiconductor materials. Notably, graphene and carbon nanofibers (CNFs) have emerged as pivotal players in the semiconductor domain, owing to their remarkable electrical, mechanical, and thermal properties, coupled with their distinctive structural attributes.

Graphene, characterized by its two-dimensional single-layer structure of densely packed carbon atoms, boasts unparalleled electrical conductivity. This positions it for significant potential in applications like highfrequency electronic devices and sensors. Furthermore, its transparency and flexibility pave the way for innovative advancements in flexible electronic devices and display technologies, rejuvenating the electronics industry’s potential. CNFs, celebrated for their nanoscale diameter and exceptional mechanical attributes, carve a niche for themselves in material science. Their superior conductivity heralds vast opportunities, especially in realms such as conductive fibers and flexible circuitry. Within the spectrum of synthesis techniques, Chemical Vapor Deposition (CVD) emerges as a standout method, particularly for producing highquality graphene films and CNFs.

This dissertation delves into the CVD preparation, performance characterization, and subsequent applications of these materials, particularly in electromagnetic (EM) and ultraviolet (UV) optics. Specifically, the research encompasses:

1. Investigating the photoluminescence (PL) performance of ZnO/graphene composite films and ZnO@CNFs composite fibers within the UV emission spectrum
2. Determining the optimal thickness of ZnO in the ZnO/graphene composite for optimal UV-PL behavior
3. Analyzing the anisotropic EM loss behavior of CNFs and ZnO@CNFs samples across the MMW and sub-terahertz bands
4. Pioneering a novel quality-assessment method for wafer-scale CVDprepared graphene films, grounded in their distinctive interactions with terahertz TE10 waves

In essence, this research centers on graphene and CNFs, exploring their potential in the realm of EM and UV optics and offering insights based on their intrinsic properties.

On June 20th 2024 at 16:00, Yangxintong Lyu will defend their PhD entitled “DEEP-LEARNING-BASED MULTI-MODAL FUSION FOR TRAFFIC IMAGE DATA PROCESSING”.

Everybody is invited to attend the presentation in room I.0.02, or digitally via this link.

In recent years, deep-learning-based technologies have significantly developed, which is driven by a large amount of data associated with task-specific labels. Among the various formats used for representing object attributes in computer vision, RGB images stand out as a ubiquitous choice. Their value extends to traffic-related applications, particularly in the realms of autonomous driving and intelligent surveillance systems. By using an autonomous driving system, a car is capable of navigating and operating with diminished human interactions, while traffic conditions can be monitored and analysed by an intelligent system. Essentially, the techniques reduce human error and improve road safety, which significantly impacts our daily life.

Although many visual-based traffic analysis tasks can indeed be effectively solved by leveraging features extracted from a sole RGB channel, certain unresolved challenges persist that introduce extra difficulties under certain situations. First of all, extracting complicated information becomes demanding, especially under erratic lighting conditions, raising a need for auxiliary clues. Secondly, obtaining large-scale accurate labels for challenging tasks remains time-consuming, costly, and arduous. The former prompts exploration into capturing and exploiting additional information such that the objects can be observed from diverse aspects; in contrast, the latter requires either an increase in the volume of available data or the capability to learn from other datasets that already possess perfect labels.

In this thesis, we tackle multi-modal data fusion and data scarcity for intelligent transportation systems. Our first contribution is a novel RGB-Thermal fusion neural network for semantic segmentation. It ensures the segmentation under limited illumination. Our second contribution is a 3D-prior-based framework for monocular vehicle 6D pose estimation. The use of 3D geometry avoids the ill-posed pose prediction from a single camera viewpoint. Thanks to the extra 3D information, our novel method can handle distant and occluded vehicles. The third contribution is a real-world, large-scale vehicle make and model dataset that contains the most popular brands operating in Europe. Moreover, we propose a two-branch deep learning vehicle make and model recognition paradigm to reduce inter-make ambiguity. The last contribution is a weakly supervised vehicle 6D pose estimation paradigm by adapting knowledge built based on a novel synthetic dataset. The dataset includes a large amount of accurate labels for vehicles. By learning from the synthetic dataset, our method allows the significant reduction of expensive real-life vehicle pose annotations.

Comprehensive experimental results reveal that the newly introduced datasets hold significant promise for deep-learning-based processing of traffic image data. Moreover, the proposed methods surpass the existing baselines in the literature. Our research not only yields high-quality scientific publications but also underscores its value across both academic and industrial domains. 

A full immersive experience of Augmented Reality for neurosurgical planning and real-time intervention, demonstrated by Taylor on the FARI immersive CAVE during the Agoria HealthTech roundtable event June 17^th , 2024.

On July 1st 2024 at 16:00, Panagiotis Gonidakis will defend their PhD entitled “DATA- AND LABEL-EFFICIENT DEEP LEARNING FOR MEDICAL IMAGE ANALYSIS APPLICATION TO LUNG NODULE DETECTION ON THORACIC CT”.

Everybody is invited to attend the presentation in room D.0.03, or digitally via this link.

Convolutional neural networks (CNNs) have been widely used to detect and classify various objects and structures in computer vision and medical imaging. Access to large sets of annotated data is commonly a prerequisite for achieving good performance. In medical imaging, acquiring adequate amounts of labelled data can often be time consuming and costly. Therefore, reducing the need for data and in particular associated annotations, is of high importance for medical imaging applications. In this work we investigated whether we can lower the need of annotated data for a supervised learning classification problem.  

We chose to tackle the problem of lung nodule detection in thoracic computed tomography (CT) imaging, as this widely investigated application allowed us to benefit from publicly available data and benchmark our methods. We designed a 3D CNN architecture to perform patch-wise classification of candidate nodules for false positive reduction. Its training, testing and fine-tuning procedure is optimized, we evaluated its performance, and we compared it with other state-of-the-art approaches in the field.  

Next, we explored how data augmentation can contribute towards more accurate and less data-demanding models. We investigated the relative benefit of increasing the amount of original data, with respect to computationally augmenting the amount of training samples. Our result indicated that in general, better performance is achieved when increasing the amount of unique data samples, or augmenting the data more extensively, as expected. Surprisingly however, we observed that after reaching a certain amount of training samples, data augmentation led to significantly better performance compared to adding unique samples. Amongst investigated augmentation methods, rotations were found to be most beneficial for improving model performance. 

Following, we investigated the benefit of combining deep learning with handcrafted features. We explored three fusion strategies with increasing complexity and assessed their performance for varying amounts of training data. Our findings indicated that combining handcrafted features with a 3D CNN approach significantly improved lung nodule detection performance in comparison to an independently trained CNN model, regardless of the fusion strategy. Comparatively larger increases in performance were obtained when less training data was available. The fusion strategy in which features are combined with a CNN using a single end-to-end training scheme performed best overall, allowing to reduce training data by 33% to 43%, while maintaining performance. Among the investigated handcrafted features, those that describe the relative position of the candidate with respect to the lung wall and mediastinum, were found to be of most benefit. 

Finally, we considered the case in which abundant data is available, but annotations are scarce, and investigated several methods to improve label-efficiency and their combined effect. We proposed a framework that utilizes both annotated and unannotated data, can be pretrained via self-supervision, and allows to combine handcrafted features with learned representations. Interestingly, the improvements in performance derived from the proposed learning schemes were found to accumulate, leading to increased label-efficiency when these strategies are combined. We observed a potential to decrease the amount of annotated data up to 68% when compared to traditional supervised training, while maintaining performance.  

Our findings indicate that the investigated methods allow considerable reduction of data and/or annotations while maintaining model performance for lung nodule detection from CT imaging. Future work should investigate whether these results generalize to other domains, such that more applications that face challenges due to a shortage of annotated data may benefit from the potential of deep learning.

On June 13th 2024 at 16:00, Remco Royen will defend their PhD entitled “ADDRESSING LABELLING, COMPLEXITY, LATENCY, AND SCALABILITY IN DEEP LEARNING-BASED PROCESSING OF POINT CLOUDS”.

Everybody is invited to attend the presentation in room I.0.01, or digitally via this link.

The field of 3D technology is attracting considerable academic and industrial interest due to its expanding range of potential applications. Noteworthy domains include, but extend beyond, automotive, gaming, extended and augmented reality, drone inspection, robotics, medical imaging, and 3D modeling and design. An essential aspect for these applications is 3D scene understanding. Point clouds, consisting of a collection of points, play a crucial role in capturing the spatial information of physical environments. This results in a lightweight and precise 3D representation, preserving fine details and enabling efficient integration with real-world data.

In recent years, deep learning has gained widespread use, demonstrating its significance across various domains. Its ability to automatically learn intricate patterns from vast datasets has resulted in a transformative impact, driving advancements in technology, and reshaping the landscape of artificial intelligence applications. The ongoing development of increasingly sophisticated neural network architectures continues to push the boundaries of what is achievable across diverse sectors.

As a result, deep learning has become ubiquitous. However, certain limitations hinder its broad applicability. This thesis delves into four crucial challenges associated with deep learning-based point cloud processing: (i) the precise labeling of extensive datasets, (ii) the model complexity requirements, (iii) the latency introduced during inference, and (iv) the concept of scalability. The initial challenge stems from the necessity for extensive datasets with highly accurate annotations. Particularly in the 3D domain, obtaining such high-quality annotations proves challenging and, consequently, expensive. The second challenged arises from the development of more intricate and memory-intensive, facilitated by advancements in high-power-consuming graphics cards. While these methods achieve higher performance levels, they impose constraints on deployment, particularly for embedded devices. Furthermore, the escalating complexity of these networks is accompanied by an increased inference time, impeding real-time applications. Lastly, deep learning-based solutions lack the concept of scalability which have proven vital in traditional methods.

In this thesis, we tackle these challenges and propose diverse solutions within the deep learning paradigm. The thesis commences with the introduction of a rapid 3D LiDAR simulator, designed to mitigate the labeling problem by learning from perfectly annotated synthetic data. We demonstrate its applications in 3D denoising and semantic segmentation. A second contribution can be found within the domain of point cloud instance segmentation. Through the joint learning of prototypes and coefficients, we present an efficient and rapid method that demands relatively low GPU memory. To further improve our method, we introduce an enhanced block merging algorithm. As a third main contribution, we achieve deep learning-based quality scalability by learning embedded latent representations, demonstrating compelling results in applications such as image reconstruction, point cloud compression, and image semantic hashing. The final contribution introduces resolution-scalable 3D semantic segmentation of point clouds. When applied to resolutionscalable 3D sensors, it enables joint point cloud acquisition and processing.

Our proposed methods consistently outperform established benchmarks across diverse datasets, as demonstrated through comprehensive experimentation. The research findings have been disseminated in various reputable journals and conferences, and have led to a patent submission, highlighting their impact in both academic and industrial contexts.

On June 17th 2024 at 10:00, Ruben De Smet will defend their PhD entitled “RAPID PROTOYPING AND DEPLOYMENT OF PRIVACY-ENHANCING TECHNOLOGIES”.

Everybody is invited to attend the presentation in room D.2.01, or digitally via this link.

Since its inception, the internet has quickly become a public service utility. The combination of its commercial exploitation, and the rather intimate nature of how humans actively use the internet, gives rise to some paradoxical situations. As a citizen of Belgium, I would probably not expect to give my name and phone number to a company in the United States to talk to my brother, 50 km up north. However, for over two billion people, this is their rather paradoxical reality: the company Meta, owning WhatsApp, collects and stores these data for their users. This cherry-picked scenario stands example for a wider trend in the industry.

Cryptographers have worked on several privacy-enhancing technologies (PETs). These PETs aim to minimize the amount of personal data to fulfil a service for users. Although these technologies exist on paper, several practical issues arise. These practicalities are the subject of this thesis.

One practical issue is the performance. PETs that run on end-user devices should both be fast and require little bandwidth. We investigate how implementation details may lead to significant speedups or bandwidth savings. Specifically, we devise a zero-knowledge proof (ZKP) tailored to electronic road pricing (ERP). ERP is a privacy-sensitive topic, and our ERP system achieves some notable performance improvements over preexisting proposals.

A second practical issue is the challenging nature of implementing PETs. We present “Circuitree” and “Whisperfish”, to study how to bring PETs to an actual application. Circuitree is a high-level framework to tailor ZKPs to specific scenarios, using a bespoke logic programming language. The language is designed such that the resulting ZKP is highly efficient.

Whisperfish is effectively a reimplementation of the Signal instant messaging client, and allows us to present in detail how Signal deploys their PETs to users. All ideas put forward in this thesis were evaluated by means of their implementation in the Rust programming language.

… Nvidia is the undisputed leader in GPUs, the processors capable of handling very complex calculations. Nvidia currently has a significant amount of cash on hand, which is useful for making substantial investments and staying ahead….

https://www.tijd.be/ondernemen/technologie/houdt-momentum-aan-voor-nvidia-pas-als-technologie-radicaal-verandert-kan-de-chipreus-in-de-problemen-komen/10547908.html

https://flux50.com/news-events/events/15de-energiecongres

Johan Stiens is a guest speaker at this 15^th Energy Congress, entitled CLIMATE-FIT by innovation iN BUILDING CONSTRUCTION on May 30^th in Mechelen, with the invited talk: INTOWALL- radar technology for inspection of the building envelope. 

On June 6th 2024 at 16:00, Anirudh Praveen Kankuppe Raghavendra Swamy will defend their PhD entitled “LOW POWER MM-WAVE FMCW RADAR RECEIVERS IN V AND D BANDS”.

Everybody is invited to attend the presentation in room D.2.01, or digitally via this link.

From the beginning of 20th Century, RADARs have been the cornerstone of the military arsenal. Radar has served as a prime sensing and tracking device and has evolved in complexity over time, while relying on simple modulation like FMCW (frequency-modulated continuous wave). Earlier radars started with operating frequencies of a few MHz, while modern radars operate at frequencies higher than 30 GHz, termed as mm-wave, to benefit from a large available bandwidth. While retaining their fundamental aspect of range and velocity sensing, today’s radars are found in diverse domestic areas like automotive, indoor sensing, human machine interface and vital signs monitoring. Though range and velocity sensing can be achieved with a single radar transceiver, to sense any angle of arrival, they must be arranged as MIMO (multiple input, multiple output) arrays. As the array size grows, it is paramount to have energy efficient array elements that do not compromise performance. The receiver of such an array should be resistant to leakage from the radar transmitter to the receiver termed as spillover. Otherwise, this spillover can result in saturation of the front-end or baseband circuitry in the receiver. Further, to have a better range resolution, the bandwidth of the front-end should be as high as possible. In this work, such a mm-wave radar receiver is explored with an emphasis on low power consumption, large RF bandwidth, robustness to spillover, and unique narrow-band filter for spillover or nearby large target attenuation. Two receivers were designed in a 28 nm bulk CMOS process, operating at 60 GHz (V-band) and 140 GHz (D- band) with a power consumption of 5.2 mW and 67 mW and a bandwidth of 10.2 GHz and 18.3 GHz, respectively. Core of the innovations aiding these state-of-the-art power consumption numbers are a mixer-first front-end architecture, a source-degenerated high-pass filter, a variable gain band-pass Gm-C filter, a low-power broadband I/Q RF front-end at D-band, and a unique, tunable narrow-band spillover and target attenuation filters. With this record low power consumption, the radars have been demonstrated to detect multi-targets, pedestrian movement, heartbeat and could filter selective targets in the range with a 13 mm range resolution marking a spot among state-of-the-art FMCW radar receivers and setting a benchmark for the future.

On May 30th 2024 at 15:00, Taylor Frantz will defend their PhD entitled “AUGMENTED REALITY IN SURGERY ON THE DEVELOPMENT OF REAL-TIME INTERVENTIONAL PLANNING AND NAVIGATION FOR NEUROSURGICAL AND ORTHOPEDIC USE CASES: BENCH-TOP TO CLINICAL EVALUATION”.

Everybody is invited to attend the presentation in room D.2.01, or digitally via this link.

Computer aided navigation (CAN) is a surgical technology which allows a surgeon to use patient medical image data as map to guide the procedure. It comprises several interconnected processes: visualization of 3D medical image data, tracking of surgical instrumentation, definition of a virtual coordinate system around the patient, and the alignment of the image data to the patient.

Despite quantitative benefits, the technology is often not used due to size, cost, and unintuitive visualization of 3D patient data as 2D black and white image. Augmented reality (AR) devices often integrate requisite hardware for CAN into a compact and mobile head mounted device (HMD) and allow the surgeon to view complex 3D data as a hologram overlying the patient. This work addresses technical limitations of such low-cost AR hardware with respect to tracking performance and presents evidence supporting their use in both neurosurgical and orthopedic domains.

Building on early work in videometric tracking as a proof-of-concept, the development of monocular infrared (IR) tracking of existing surgical instrumentation provided a method to establish a room-stable coordinate system and a mechanism for precise user input; both required for CAN. This tracking solution was validated in a VICON motion capture lab and demonstrated a mean pose estimation error of 0.78 mm ± 0.74 mm and 0.84° ± 0.64°.

Following this, phantom trials in navigated external ventricular drain (EVD) placement, and total shoulder and hip arthroplasty were performed. The results demonstrated a reduction in technique learning curve of the former, and improved outcomes of the latter when compared to traditional non-navigated techniques. Moreover, AR data registration was found to be comparable to modern CAN systems.

Clinical trials in both tumor resection planning and EVD were then performed to assess efficacy of AR-CAN compared to current surgical practice. In the former, AR-CAN demonstrated a reduction in preoperative planning time with superior lesion delineation when compared to neuronavigation. Preliminary results in AR navigated EVD placement outcomes demonstrate 82% optimal (grade I), 18% sub optimal (grade II), and 0% (grade III). This currently outperforms literature, given single attempt insertion.

“Signal Processing in the AI era” was the tagline of this year’s IEEE International Conference on Acoustics, Speech and Signal Processing, taking place in Rhodes, Greece.

In this context, Brent de Weerdt, Xiangyu Yang, Boris Joukovsky, Alex Stergiou and Nikos Deligiannis presented ETRO’s research during poster sessions and oral presentations, with novel ways to process and understand graph, video, and audio data. Nikos Deligiannis chaired a session on Graph Deep Learning, attended the IEEE T-IP Editorial Board Meeting, and had the opportunity to meet with collaborators from the VUB-Duke-Ugent-UCL joint lab.

Featured articles:

• De Weerdt, B. et al., “Designing Transformer Networks for Sparse Recovery of Sequential Data Using Deep Unfolding” (https://ieeexplore.ieee.org/abstract/document/10094712)
• Yang X. et al., “Relevance Propagation through Deep Conditional Random Fields” (https://ieeexplore.ieee.org/abstract/document/10095075)
• Stergiou A. et al., “Play it back: Iterative attention for audio recognition” (https://ieeexplore.ieee.org/abstract/document/10096532, developed at the University of Bristol)

The paper entitled” Novel Multi-Parametric Sensor System for Comprehensive Multi-Wavelength Photoplethysmography Characterization”, published in the journal MDPI SENSORS in 2023, coauthored by: J.L. Cause, Á. Solé-Morillo, B. Da Silva, J. García-Naranjo, and J. Stiens was awarded with the first prize of the provincial council of scientific health societies of Santiago de Cuba

Johan Stiens gave an interview with deputy of the province Flemish-Brabant on Ring TV about the TRN radar in support for renovation in the construction industry

Subsidies van 600.000 euro voor innoverende bedrijven in Vlaams-Brabant | Ring (ringtv.be)

600.000 euro voor innoverende Vlaams-Brabantse bedrijven | Voka

On June 27th 2024 at 09:30, Cheng Chen will defend their PhD entitled “NOVEL FABRICATION AND ELECTROMAGNETIC-OPTICAL CHARACTERIZATION TECHNIQUES OF CARBON-BASED NANOSTRUCTURES”.

Everybody is invited to attend the presentation in room D.2.01, or digitally via this link.

Nowadays, carbon nanomaterials are increasingly garnering attention as the next generation of semiconductor materials. Notably, graphene and carbon nanofibers (CNFs) have emerged as pivotal players in the semiconductor domain, owing to their remarkable electrical, mechanical, and thermal properties, coupled with their distinctive structural attributes.

Graphene, characterized by its two-dimensional single-layer structure of densely packed carbon atoms, boasts unparalleled electrical conductivity. This positions it for significant potential in applications like highfrequency electronic devices and sensors. Furthermore, its transparency and flexibility pave the way for innovative advancements in flexible electronic devices and display technologies, rejuvenating the electronics industry’s potential. CNFs, celebrated for their nanoscale diameter and exceptional mechanical attributes, carve a niche for themselves in material science. Their superior conductivity heralds vast opportunities, especially in realms such as conductive fibers and flexible circuitry. Within the spectrum of synthesis techniques, Chemical Vapor Deposition (CVD) emerges as a standout method, particularly for producing highquality graphene films and CNFs.

This dissertation delves into the CVD preparation, performance characterization, and subsequent applications of these materials, particularly in electromagnetic (EM) and ultraviolet (UV) optics. Specifically, the research encompasses:

1. Investigating the photoluminescence (PL) performance of ZnO/graphene composite films and ZnO@CNFs composite fibers within the UV emission spectrum
2. Determining the optimal thickness of ZnO in the ZnO/graphene composite for optimal UV-PL behavior
3. Analyzing the anisotropic EM loss behavior of CNFs and ZnO@CNFs samples across the MMW and sub-terahertz bands
4. Pioneering a novel quality-assessment method for wafer-scale CVDprepared graphene films, grounded in their distinctive interactions with terahertz TE10 waves

In essence, this research centers on graphene and CNFs, exploring their potential in the realm of EM and UV optics and offering insights based on their intrinsic properties.

On June 20th 2024 at 16:00, Yangxintong Lyu will defend their PhD entitled “DEEP-LEARNING-BASED MULTI-MODAL FUSION FOR TRAFFIC IMAGE DATA PROCESSING”.

Everybody is invited to attend the presentation in room I.0.02, or digitally via this link.

In recent years, deep-learning-based technologies have significantly developed, which is driven by a large amount of data associated with task-specific labels. Among the various formats used for representing object attributes in computer vision, RGB images stand out as a ubiquitous choice. Their value extends to traffic-related applications, particularly in the realms of autonomous driving and intelligent surveillance systems. By using an autonomous driving system, a car is capable of navigating and operating with diminished human interactions, while traffic conditions can be monitored and analysed by an intelligent system. Essentially, the techniques reduce human error and improve road safety, which significantly impacts our daily life.

Although many visual-based traffic analysis tasks can indeed be effectively solved by leveraging features extracted from a sole RGB channel, certain unresolved challenges persist that introduce extra difficulties under certain situations. First of all, extracting complicated information becomes demanding, especially under erratic lighting conditions, raising a need for auxiliary clues. Secondly, obtaining large-scale accurate labels for challenging tasks remains time-consuming, costly, and arduous. The former prompts exploration into capturing and exploiting additional information such that the objects can be observed from diverse aspects; in contrast, the latter requires either an increase in the volume of available data or the capability to learn from other datasets that already possess perfect labels.

In this thesis, we tackle multi-modal data fusion and data scarcity for intelligent transportation systems. Our first contribution is a novel RGB-Thermal fusion neural network for semantic segmentation. It ensures the segmentation under limited illumination. Our second contribution is a 3D-prior-based framework for monocular vehicle 6D pose estimation. The use of 3D geometry avoids the ill-posed pose prediction from a single camera viewpoint. Thanks to the extra 3D information, our novel method can handle distant and occluded vehicles. The third contribution is a real-world, large-scale vehicle make and model dataset that contains the most popular brands operating in Europe. Moreover, we propose a two-branch deep learning vehicle make and model recognition paradigm to reduce inter-make ambiguity. The last contribution is a weakly supervised vehicle 6D pose estimation paradigm by adapting knowledge built based on a novel synthetic dataset. The dataset includes a large amount of accurate labels for vehicles. By learning from the synthetic dataset, our method allows the significant reduction of expensive real-life vehicle pose annotations.

Comprehensive experimental results reveal that the newly introduced datasets hold significant promise for deep-learning-based processing of traffic image data. Moreover, the proposed methods surpass the existing baselines in the literature. Our research not only yields high-quality scientific publications but also underscores its value across both academic and industrial domains. 

A full immersive experience of Augmented Reality for neurosurgical planning and real-time intervention, demonstrated by Taylor on the FARI immersive CAVE during the Agoria HealthTech roundtable event June 17^th , 2024.

On July 1st 2024 at 16:00, Panagiotis Gonidakis will defend their PhD entitled “DATA- AND LABEL-EFFICIENT DEEP LEARNING FOR MEDICAL IMAGE ANALYSIS APPLICATION TO LUNG NODULE DETECTION ON THORACIC CT”.

Everybody is invited to attend the presentation in room D.0.03, or digitally via this link.

Convolutional neural networks (CNNs) have been widely used to detect and classify various objects and structures in computer vision and medical imaging. Access to large sets of annotated data is commonly a prerequisite for achieving good performance. In medical imaging, acquiring adequate amounts of labelled data can often be time consuming and costly. Therefore, reducing the need for data and in particular associated annotations, is of high importance for medical imaging applications. In this work we investigated whether we can lower the need of annotated data for a supervised learning classification problem.  

We chose to tackle the problem of lung nodule detection in thoracic computed tomography (CT) imaging, as this widely investigated application allowed us to benefit from publicly available data and benchmark our methods. We designed a 3D CNN architecture to perform patch-wise classification of candidate nodules for false positive reduction. Its training, testing and fine-tuning procedure is optimized, we evaluated its performance, and we compared it with other state-of-the-art approaches in the field.  

Next, we explored how data augmentation can contribute towards more accurate and less data-demanding models. We investigated the relative benefit of increasing the amount of original data, with respect to computationally augmenting the amount of training samples. Our result indicated that in general, better performance is achieved when increasing the amount of unique data samples, or augmenting the data more extensively, as expected. Surprisingly however, we observed that after reaching a certain amount of training samples, data augmentation led to significantly better performance compared to adding unique samples. Amongst investigated augmentation methods, rotations were found to be most beneficial for improving model performance. 

Following, we investigated the benefit of combining deep learning with handcrafted features. We explored three fusion strategies with increasing complexity and assessed their performance for varying amounts of training data. Our findings indicated that combining handcrafted features with a 3D CNN approach significantly improved lung nodule detection performance in comparison to an independently trained CNN model, regardless of the fusion strategy. Comparatively larger increases in performance were obtained when less training data was available. The fusion strategy in which features are combined with a CNN using a single end-to-end training scheme performed best overall, allowing to reduce training data by 33% to 43%, while maintaining performance. Among the investigated handcrafted features, those that describe the relative position of the candidate with respect to the lung wall and mediastinum, were found to be of most benefit. 

Finally, we considered the case in which abundant data is available, but annotations are scarce, and investigated several methods to improve label-efficiency and their combined effect. We proposed a framework that utilizes both annotated and unannotated data, can be pretrained via self-supervision, and allows to combine handcrafted features with learned representations. Interestingly, the improvements in performance derived from the proposed learning schemes were found to accumulate, leading to increased label-efficiency when these strategies are combined. We observed a potential to decrease the amount of annotated data up to 68% when compared to traditional supervised training, while maintaining performance.  

Our findings indicate that the investigated methods allow considerable reduction of data and/or annotations while maintaining model performance for lung nodule detection from CT imaging. Future work should investigate whether these results generalize to other domains, such that more applications that face challenges due to a shortage of annotated data may benefit from the potential of deep learning.

On June 13th 2024 at 16:00, Remco Royen will defend their PhD entitled “ADDRESSING LABELLING, COMPLEXITY, LATENCY, AND SCALABILITY IN DEEP LEARNING-BASED PROCESSING OF POINT CLOUDS”.

Everybody is invited to attend the presentation in room I.0.01, or digitally via this link.

The field of 3D technology is attracting considerable academic and industrial interest due to its expanding range of potential applications. Noteworthy domains include, but extend beyond, automotive, gaming, extended and augmented reality, drone inspection, robotics, medical imaging, and 3D modeling and design. An essential aspect for these applications is 3D scene understanding. Point clouds, consisting of a collection of points, play a crucial role in capturing the spatial information of physical environments. This results in a lightweight and precise 3D representation, preserving fine details and enabling efficient integration with real-world data.

In recent years, deep learning has gained widespread use, demonstrating its significance across various domains. Its ability to automatically learn intricate patterns from vast datasets has resulted in a transformative impact, driving advancements in technology, and reshaping the landscape of artificial intelligence applications. The ongoing development of increasingly sophisticated neural network architectures continues to push the boundaries of what is achievable across diverse sectors.

As a result, deep learning has become ubiquitous. However, certain limitations hinder its broad applicability. This thesis delves into four crucial challenges associated with deep learning-based point cloud processing: (i) the precise labeling of extensive datasets, (ii) the model complexity requirements, (iii) the latency introduced during inference, and (iv) the concept of scalability. The initial challenge stems from the necessity for extensive datasets with highly accurate annotations. Particularly in the 3D domain, obtaining such high-quality annotations proves challenging and, consequently, expensive. The second challenged arises from the development of more intricate and memory-intensive, facilitated by advancements in high-power-consuming graphics cards. While these methods achieve higher performance levels, they impose constraints on deployment, particularly for embedded devices. Furthermore, the escalating complexity of these networks is accompanied by an increased inference time, impeding real-time applications. Lastly, deep learning-based solutions lack the concept of scalability which have proven vital in traditional methods.

In this thesis, we tackle these challenges and propose diverse solutions within the deep learning paradigm. The thesis commences with the introduction of a rapid 3D LiDAR simulator, designed to mitigate the labeling problem by learning from perfectly annotated synthetic data. We demonstrate its applications in 3D denoising and semantic segmentation. A second contribution can be found within the domain of point cloud instance segmentation. Through the joint learning of prototypes and coefficients, we present an efficient and rapid method that demands relatively low GPU memory. To further improve our method, we introduce an enhanced block merging algorithm. As a third main contribution, we achieve deep learning-based quality scalability by learning embedded latent representations, demonstrating compelling results in applications such as image reconstruction, point cloud compression, and image semantic hashing. The final contribution introduces resolution-scalable 3D semantic segmentation of point clouds. When applied to resolutionscalable 3D sensors, it enables joint point cloud acquisition and processing.

Our proposed methods consistently outperform established benchmarks across diverse datasets, as demonstrated through comprehensive experimentation. The research findings have been disseminated in various reputable journals and conferences, and have led to a patent submission, highlighting their impact in both academic and industrial contexts.

On June 17th 2024 at 10:00, Ruben De Smet will defend their PhD entitled “RAPID PROTOYPING AND DEPLOYMENT OF PRIVACY-ENHANCING TECHNOLOGIES”.

Everybody is invited to attend the presentation in room D.2.01, or digitally via this link.

Since its inception, the internet has quickly become a public service utility. The combination of its commercial exploitation, and the rather intimate nature of how humans actively use the internet, gives rise to some paradoxical situations. As a citizen of Belgium, I would probably not expect to give my name and phone number to a company in the United States to talk to my brother, 50 km up north. However, for over two billion people, this is their rather paradoxical reality: the company Meta, owning WhatsApp, collects and stores these data for their users. This cherry-picked scenario stands example for a wider trend in the industry.

Cryptographers have worked on several privacy-enhancing technologies (PETs). These PETs aim to minimize the amount of personal data to fulfil a service for users. Although these technologies exist on paper, several practical issues arise. These practicalities are the subject of this thesis.

One practical issue is the performance. PETs that run on end-user devices should both be fast and require little bandwidth. We investigate how implementation details may lead to significant speedups or bandwidth savings. Specifically, we devise a zero-knowledge proof (ZKP) tailored to electronic road pricing (ERP). ERP is a privacy-sensitive topic, and our ERP system achieves some notable performance improvements over preexisting proposals.

A second practical issue is the challenging nature of implementing PETs. We present “Circuitree” and “Whisperfish”, to study how to bring PETs to an actual application. Circuitree is a high-level framework to tailor ZKPs to specific scenarios, using a bespoke logic programming language. The language is designed such that the resulting ZKP is highly efficient.

Whisperfish is effectively a reimplementation of the Signal instant messaging client, and allows us to present in detail how Signal deploys their PETs to users. All ideas put forward in this thesis were evaluated by means of their implementation in the Rust programming language.

… Nvidia is the undisputed leader in GPUs, the processors capable of handling very complex calculations. Nvidia currently has a significant amount of cash on hand, which is useful for making substantial investments and staying ahead….

https://www.tijd.be/ondernemen/technologie/houdt-momentum-aan-voor-nvidia-pas-als-technologie-radicaal-verandert-kan-de-chipreus-in-de-problemen-komen/10547908.html

https://flux50.com/news-events/events/15de-energiecongres

Johan Stiens is a guest speaker at this 15^th Energy Congress, entitled CLIMATE-FIT by innovation iN BUILDING CONSTRUCTION on May 30^th in Mechelen, with the invited talk: INTOWALL- radar technology for inspection of the building envelope. 

On June 6th 2024 at 16:00, Anirudh Praveen Kankuppe Raghavendra Swamy will defend their PhD entitled “LOW POWER MM-WAVE FMCW RADAR RECEIVERS IN V AND D BANDS”.

Everybody is invited to attend the presentation in room D.2.01, or digitally via this link.

From the beginning of 20th Century, RADARs have been the cornerstone of the military arsenal. Radar has served as a prime sensing and tracking device and has evolved in complexity over time, while relying on simple modulation like FMCW (frequency-modulated continuous wave). Earlier radars started with operating frequencies of a few MHz, while modern radars operate at frequencies higher than 30 GHz, termed as mm-wave, to benefit from a large available bandwidth. While retaining their fundamental aspect of range and velocity sensing, today’s radars are found in diverse domestic areas like automotive, indoor sensing, human machine interface and vital signs monitoring. Though range and velocity sensing can be achieved with a single radar transceiver, to sense any angle of arrival, they must be arranged as MIMO (multiple input, multiple output) arrays. As the array size grows, it is paramount to have energy efficient array elements that do not compromise performance. The receiver of such an array should be resistant to leakage from the radar transmitter to the receiver termed as spillover. Otherwise, this spillover can result in saturation of the front-end or baseband circuitry in the receiver. Further, to have a better range resolution, the bandwidth of the front-end should be as high as possible. In this work, such a mm-wave radar receiver is explored with an emphasis on low power consumption, large RF bandwidth, robustness to spillover, and unique narrow-band filter for spillover or nearby large target attenuation. Two receivers were designed in a 28 nm bulk CMOS process, operating at 60 GHz (V-band) and 140 GHz (D- band) with a power consumption of 5.2 mW and 67 mW and a bandwidth of 10.2 GHz and 18.3 GHz, respectively. Core of the innovations aiding these state-of-the-art power consumption numbers are a mixer-first front-end architecture, a source-degenerated high-pass filter, a variable gain band-pass Gm-C filter, a low-power broadband I/Q RF front-end at D-band, and a unique, tunable narrow-band spillover and target attenuation filters. With this record low power consumption, the radars have been demonstrated to detect multi-targets, pedestrian movement, heartbeat and could filter selective targets in the range with a 13 mm range resolution marking a spot among state-of-the-art FMCW radar receivers and setting a benchmark for the future.

On May 30th 2024 at 15:00, Taylor Frantz will defend their PhD entitled “AUGMENTED REALITY IN SURGERY ON THE DEVELOPMENT OF REAL-TIME INTERVENTIONAL PLANNING AND NAVIGATION FOR NEUROSURGICAL AND ORTHOPEDIC USE CASES: BENCH-TOP TO CLINICAL EVALUATION”.

Everybody is invited to attend the presentation in room D.2.01, or digitally via this link.

Computer aided navigation (CAN) is a surgical technology which allows a surgeon to use patient medical image data as map to guide the procedure. It comprises several interconnected processes: visualization of 3D medical image data, tracking of surgical instrumentation, definition of a virtual coordinate system around the patient, and the alignment of the image data to the patient.

Despite quantitative benefits, the technology is often not used due to size, cost, and unintuitive visualization of 3D patient data as 2D black and white image. Augmented reality (AR) devices often integrate requisite hardware for CAN into a compact and mobile head mounted device (HMD) and allow the surgeon to view complex 3D data as a hologram overlying the patient. This work addresses technical limitations of such low-cost AR hardware with respect to tracking performance and presents evidence supporting their use in both neurosurgical and orthopedic domains.

Building on early work in videometric tracking as a proof-of-concept, the development of monocular infrared (IR) tracking of existing surgical instrumentation provided a method to establish a room-stable coordinate system and a mechanism for precise user input; both required for CAN. This tracking solution was validated in a VICON motion capture lab and demonstrated a mean pose estimation error of 0.78 mm ± 0.74 mm and 0.84° ± 0.64°.

Following this, phantom trials in navigated external ventricular drain (EVD) placement, and total shoulder and hip arthroplasty were performed. The results demonstrated a reduction in technique learning curve of the former, and improved outcomes of the latter when compared to traditional non-navigated techniques. Moreover, AR data registration was found to be comparable to modern CAN systems.

Clinical trials in both tumor resection planning and EVD were then performed to assess efficacy of AR-CAN compared to current surgical practice. In the former, AR-CAN demonstrated a reduction in preoperative planning time with superior lesion delineation when compared to neuronavigation. Preliminary results in AR navigated EVD placement outcomes demonstrate 82% optimal (grade I), 18% sub optimal (grade II), and 0% (grade III). This currently outperforms literature, given single attempt insertion.