Abstract The segmentation and fitting of geometric primitives from point clouds is a widely adopted approach for modelling the underlying geometric structure of objects in reverse engineering and numerous graphics applications. Existing methods either overlook the role of geometric information in assisting segmentation or incorporate reconstruction losses without leveraging modern neural implicit field representations, leading to limited robustness against noise and weak expressive power in reconstruction. We propose a point cloud segmentation and fitting framework based on neural implicit representations, fully leveraging neural implicit fields' expressive power and robustness. The key idea is the unification of geometric representation within a neural implicit field framework, enabling seamless integration of geometric loss for improved performance. In contrast to previous approaches that focus solely on clustering in the feature embedding space, our method enhances instance segmentation through semantic‐aware point embeddings and simultaneously improves semantic predictions via instance‐level feature fusion. Furthermore, we incorporate 3D‐specific cues such as spatial dimensions and geometric connectivity, which are uniquely informative in the 3D domain. Extensive experiments and comparisons against previous methods demonstrate our robustness and superiority.

In this paper, we introduce an innovative method for computer-aided design (CAD) segmentation by concatenating meshes and CAD models. Many previous CAD segmentation methods have achieved impressive performance using single representations, such as meshes, CAD, and point clouds. However, existing methods cannot effectively combine different three-dimensional model types for the direct conversion, alignment, and integrity maintenance of geometric and topological information. Hence, we propose an integration approach that combines the geometric accuracy of CAD data with the flexibility of mesh representations, as well as introduce a unique hybrid representation that combines CAD and mesh models to enhance segmentation accuracy. To combine these two model types, our hybrid system utilizes advanced-neural-network techniques to convert CAD models into mesh models. For complex CAD models, model segmentation is crucial for model retrieval and reuse. In partial retrieval, it aims to segment a complex CAD model into several simple components. The first component of our hybrid system involves advanced mesh-labeling algorithms that harness the digitization of CAD properties to mesh models. The second component integrates labelled face features for CAD segmentation by leveraging the abundant multisemantic information embedded in CAD models. This combination of mesh and CAD not only refines the accuracy of boundary delineation but also provides a comprehensive understanding of the underlying object semantics. This study uses the Fusion 360 Gallery dataset. Experimental results indicate that our hybrid method can segment these models with higher accuracy than other methods that use single representations.

Computer-Aided Design (CAD) plays a crucial role in industrial manufacturing by providing geometry information and the construction workflow for manufactured objects. The construction information enables effective re-editing of parametric CAD models. While boundary representation (B-Rep) is the standard format for representing geometry structures, JSON format is an alternative due to the lack of uniform criteria for storing the construction workflow. Regrettably, most CAD models available on the Internet only offer geometry information, omitting the construction procedure and hampering creation efficiency. This paper proposes a learning approach CADParser to infer the underlying modeling sequences given a B-Rep CAD model. It achieves this by treating the CAD geometry structure as a graph and the construction workflow as a sequence. Since the existing CAD dataset only contains two operations (i.e., Sketch and Extrusion), limiting the diversity of the CAD model creation, we also introduce a large-scale dataset incorporating a more comprehensive range of operations such as Revolution, Fillet, and Chamfer. Each model includes both the geometry structure and the construction sequences. Extensive experiments demonstrate that our method can compete with the existing state-of-the-art methods quantitatively and qualitatively. Data is available at https://drive.google.com/CADParserData.

Existing enhancement methods are empirically expected to help the high-level end computer vision task: however, that is observed to not always be the case in practice. We focus on object or face detection in poor visibility enhancements caused by bad weathers (haze, rain) and low light conditions. To provide a more thorough examination and fair comparison, we introduce three benchmark sets collected in real-world hazy, rainy, and low-light conditions, respectively, with annotated objects/faces. We launched the UG2+ challenge Track 2 competition in IEEE CVPR 2019, aiming to evoke a comprehensive discussion and exploration about whether and how low-level vision techniques can benefit the high-level automatic visual recognition in various scenarios. To our best knowledge, this is the first and currently largest effort of its kind. Baseline results by cascading existing enhancement and detection models are reported, indicating the highly challenging nature of our new data as well as the large room for further technical innovations. Thanks to a large participation from the research community, we are able to analyze representative team solutions, striving to better identify the strengths and limitations of existing mindsets as well as the future directions.

Capturing images under high ISO mode introduces much noise. The statistics of high ISO noise is quite different from that of Gaussian noise. Therefore, this kind of noise is difficult to be removed by traditional Gaussian noise removal methods. This paper proposes a convolutional neural network (CNN) based method to jointly estimate and remove high ISO noise. There are two contributions in this paper. First, we propose a CNN based noise estimation method to estimate the pixel-wise noise level. Due to the Bayer down-sampling process in imaging, the noise variance map is characterized by Bayer patterns. Therefore, we propose packing 2 × 2 blocks in a noisy image into 4D vectors, which makes the pixels with similar noise levels be neighbors. Second, the noise variance map is correlated with the image content. Thus, we propose concatenating the estimated noise variance map with the noisy image, and feed the fused data to the denoising network. The two networks are trained together in an end-to-end fashion. Experimental results demonstrate that the proposed method outperforms state-of-the-art noise estimation and removal methods.