PD-Refiner: An Underlying Surface Inheritance Refiner with Adaptive Edge-Aware Supervision for Point Cloud Denoising
Abstract: Point clouds from real-world scenarios inevitably contain complex noise, significantly impairing the accuracy of downstream tasks. To tackle this challenge, cascading encoder-decoder architecture has become a conventional technical route to iterative denoise. However, circularly feeding the output of denoiser as its input again involves the re-extraction of underlying surface, leading to unstable denoising process and over-smoothed geometric details. To address these issues, we propose a novel denoising paradigm dubbed PD-Refiner that employs a single encoder to model the underlying surface. Then, we leverage several lightweight hierarchical Underlying Surface Inheritance Refiners (USIRs) to inherit and strengthen it, thereby avoiding the re-extraction from the intermediate point cloud. Furthermore, we design adaptive edge-aware supervision to improve the edge awareness of the USIRs, allowing for the adjustment of the denoising preferences from global structure to local details. The results demonstrate that our method not only achieves state-of-the-art performance in terms of denoising stability and efficacy, but also enhances edge clarity and point cloud uniformity.
Primary Subject Area: [Generation] Generative Multimedia
Secondary Subject Area: [Experience] Multimedia Applications
Relevance To Conference: The advancement of multimedia technology has triggered a new wave of imaging advancements, transitioning from traditional images/videos to the era of immersive imaging represented by virtual reality (VR), augmented reality (AR), and mixed reality (MR). In this epoch, point clouds, as a data representation method for immersive vision, are considered a reliable approach towards immersive true three-dimensional representation and rendering. They have also emerged as a research hotspot in the field of multimedia technology in recent years. Point clouds find wide applications in immersive remote rendering, autonomous driving, gaming, and other fields, all of which require high quality and accuracy from point clouds.
Point clouds are crucial in multimedia applications; however, those acquired from the real world often contain complex noise, severely impacting the performance of related downstream tasks.
To address this, we propose a novel yet practical point cloud denoising framework that boosts denoising stability by inheriting and strengthening the underlying surface representation without re-extracting it from intermediate point clouds. This methodology ensures a more stable denoising process by maintaining a consistent representation across denoising steps. We establish a stable denoising process that promotes the network to focus more on restoring geometric details. To this end, we introduce the Adaptive Shape Preserving Loss (ASPL), an adaptive supervision strategy that precisely adjusts the edge-awareness intensity of each USIR layer, ensuring detailed and accurate edge restoration. Extensive experiments demonstrate that PD-Refiner achieves superior performance compared with previous SOTA iterative methods in both testing or training phases with demanding fewer computational resources. It is noteworthy that, it also benefits the promoting of denoising quality, including the local geometric details and the uniformity. This provides strong support for creating high-quality and realistic multimedia content.
Submission Number: 4555
Loading