Advances in Point Cloud Processing and Analysis

The field of point cloud processing and analysis is rapidly evolving, with a focus on developing innovative methods for improving the robustness and accuracy of 3D deep learning models. Recent research has explored the use of novel frameworks and techniques, such as medial axis transform and diffusion models, to enhance the transferability and undefendability of point cloud attacks. Additionally, there is a growing interest in developing unified multimodal frameworks for bridging 2D and 3D industrial anomaly detection, as well as refining point cloud registration algorithms via zero-shot learning. Noteworthy papers in this area include: PDT, which presents a novel framework for point distribution transformation with diffusion models, successfully transforming input point clouds into various forms of structured outputs. RARE, which proposes a novel zero-shot method for refining point cloud registration algorithms, leveraging correspondences derived from depth images to enhance point feature representations. MAT-Adv, which introduces a novel adversarial attack framework that enhances both transferability and undefendability by explicitly perturbing the medial axis transform representations.

Sources

Transferable and Undefendable Point Cloud Attacks via Medial Axis Transform

PDT: Point Distribution Transformation with Diffusion Models

BridgeNet: A Unified Multimodal Framework for Bridging 2D and 3D Industrial Anomaly Detection

RARE: Refine Any Registration of Pairwise Point Clouds via Zero-Shot Learning

Generating Adversarial Point Clouds Using Diffusion Model

XAI for Point Cloud Data using Perturbations based on Meaningful Segmentation

Built with on top of