Thema5126: Unterschied zwischen den Versionen

Aktuelle Version vom 16. Januar 2024, 12:31 Uhr

Explainable Graph-based 3D Point Cloud Analysis

Informationen zur Arbeit

Abschlussarbeitstyp: Bachelor, Master
Betreuer: Zhan Qu
Forschungsgruppe: Web Science

Archivierungsnummer: 5126
Abschlussarbeitsstatus: Offen
Beginn: 16. Januar 2024
Abgabe: unbekannt

Weitere Informationen

The utilization of Point Cloud data has emerged as a pivotal component in the field of 3D object representation, finding relevance in various domains such as autonomous driving, robotics, drones, and augmented/virtual reality. Within the domain of 3D Point Cloud analysis, fundamental tasks include classification and segmentation. Numerous models have been developed to address these challenges, with Graph-based methods, such as Dynamic Graph Convolutional Neural Network (DGCNN) [1] and its variations, have achieved state-of-the-art performance. However, the lack of explainability limits their practical applicability [2].

The principal aim of this thesis is to develop a novel method to enhance the explainability of Graph-based models for 3D Point Cloud Analysis. By undertaking this thesis, you will not only contribute to bridging the gap between cutting-edge 3D Point Cloud analysis techniques and Graph-based models, but also develop a deep understanding of how to use explainable AI to improve the trustworthiness and interpretability of Deep Neural Networks in real-world applications.

[1] DGCNN [2] Point Cloud Saliency Maps

Ausschreibung: Download (pdf)

@@ Zeile 7: / Zeile 7: @@
 |Forschungsgruppe=Web Science
 |Abschlussarbeitsstatus=Offen
-|Beginn=2024/1/16
+|Beginn=2024/01/16
 |Ausschreibung=Thesis_Explainable3DPointCloudAnalysis.pdf
-|Beschreibung DE=Background
+|Beschreibung DE=The utilization of Point Cloud data has emerged as a pivotal component in the field of 3D object representation, finding relevance in various domains such as autonomous driving, robotics, drones, and augmented/virtual reality. Within the domain of 3D Point Cloud analysis, fundamental tasks include classification and segmentation. Numerous models have been developed to address these challenges, with Graph-based methods, such as Dynamic Graph Convolutional Neural Network (DGCNN) [1] and its variations, have achieved state-of-the-art performance. However, the lack of explainability limits their practical applicability [2].
-The utilization of Point Cloud data has emerged as a pivotal component in the field of 3D object representation, finding relevance in various domains such as autonomous driving, robotics, drones, and augmented/virtual reality. Within the domain of 3D Point Cloud analysis, fundamental tasks include classification and segmentation. Numerous models have been developed to address these challenges, with Graph-based methods, such as Dynamic Graph Convolutional Neural Network (DGCNN) [1] and its variations, have achieved state-of-the-art performance. However, the lack of explainability limits their practical applicability [2].
 The principal aim of this thesis is to develop a novel method to enhance the explainability of Graph-based models for 3D Point Cloud Analysis. By undertaking this thesis, you will not only contribute to bridging the gap between cutting-edge 3D Point Cloud analysis techniques and Graph-based models, but also develop a deep understanding of how to use explainable AI to improve the trustworthiness and interpretability of Deep Neural Networks in real-world applications.
-Prerequisites
-• Good programming skills in Python.
-• Excellent communication and academic writing skills in English.
-• Knowledge in Machine Learning, Deep Learning and Artificial Intelligence.
-• Experience with Graph Neural Networks and Explainable AI is a plus.
 [1] DGCNN
 [2] Point Cloud Saliency Maps
 }}