Abstract Computer‐Aided Design (CAD) boosts modern manufacturing, yet design reuse remains constrained by the absence of large, openly available CAD repositories with rich multi‐modal annotations suitable for search/retrieval. Recent large‐scale efforts to annotate public datasets rely on hash‐based redundancy removal that leaves no semantic structure, and on captioning by Vision‐Language Models (VLMs) using rendered images alone, which struggles to capture geometric and procedural information. We introduce MM‐CAD, a multi‐modal CAD dataset designed to level‐up retrieval and retrieval‐augmented generation models for engineering geometry, comprising two complementary parts. MM‐CAD:A brings 33,816 unique CAD models from eleven widely used benchmark datasets under a common identifier scheme, with isometric renderings, point clouds, and humanly‐curated multi‐level text captions, and 4,376 real hand‐drawn user sketches among others. MM‐CAD:B curates 192,626 models from the 1M‐model ABC corpus through a seven‐stage pipeline centered on Manifold‐Aware Adaptive Sampling (MAAS), which organizes models into semantically coherent neighborhoods rather than merely removing duplicates, directly supplying the hard negatives that contrastive retrieval training requires. Every retained model is annotated through a metadata‐grounded pipeline that conditions caption generation on parsed construction sequences rather than rendered views alone, producing three‐level text descriptions, multi‐level contour sketches, a hierarchical application taxonomy, and photorealistic in‐context images that largely preserve source CAD geometry, a modality not previously available at this scale on CAD data. We further introduce a joint retrieval architecture that aligns sketch, text, image, B‐Rep, and point cloud encoders in a single latent space through Matryoshka‐nested contrastive objectives, establishing the first unified cross‐modal retrieval benchmark for large‐scale CAD.

International audience

The success of deep learning methods led to significant breakthroughs in 3-D point cloud processing tasks with applications in remote sensing. Existing methods utilize convolutions that have some limitations, as they assume a uniform input distribution and cannot learn long-range dependences. Recent works have shown that adding attention in conjunction with these methods improves performance. This raises a question: can attention layers completely replace convolutions? This letter proposes a fully attentional model—Point Transformer (PT) for deriving a rich point cloud representation. The model’s shape classification and retrieval performance are evaluated on a large-scale urban data set—RoofN3D and a standard benchmark data set ModelNet40. Extensive experiments are conducted to test the model’s robustness to unseen point corruptions for analyzing its effectiveness on real data sets. The proposed method outperforms other state-of-the-art models in the RoofN3D data set, gives competitive results in the ModelNet40 benchmark, and shows high robustness to various unseen point corruptions. Furthermore, the model is highly memory and space-efficient when compared to other methods.

Abstract Curve reconstruction from unstructured points in a plane is a fundamental problem with many applications that has generated research interest for decades. Involved aspects like handling open, sharp, multiple and non‐manifold outlines, run‐time and provability as well as potential extension to 3D for surface reconstruction have led to many different algorithms. We survey the literature on 2D curve reconstruction and then present an open‐sourced benchmark for the experimental study. Our unprecedented evaluation of a selected set of planar curve reconstruction algorithms aims to give an overview of both quantitative analysis and qualitative aspects for helping users to select the right algorithm for specific problems in the field. Our benchmark framework is available online to permit reproducing the results and easy integration of new algorithms.

Classification of light detection and ranging (LiDAR) point clouds of building roofs plays a vital role in various urban management applications and is significant in geographic information systems (GISs) and remote sensing. In this letter, a novel deep learning-based method is proposed for classifying roof point clouds, which outperforms the state-of-the-art methods. We use a view-based method called a multiview convolutional neural network with self-attention (MVCNN-SA), which takes the multiple views of a roof point cloud as input and outputs the category of the roof. Current view-based approaches treat all views equally and simply combine the view features into a single compact 3-D descriptor. Our adaptive weight-learning algorithm, which uses the SA block, discovers the relative importance of each view, thus assigning relative weights to the views. This enhances the shape descriptor, resulting in better classification performance. The effectiveness of the proposed method is then verified on the publicly available data set - RoofN3D - by comparing it with the current state-of-the-art methods.

Abstract We present an incremental Voronoi vertex labelling algorithm for approximating contours, medial axes and dominant points (high curvature points) from 2D point sets. Though there exist many number of algorithms for reconstructing curves, medial axes or dominant points, a unified framework capable of approximating all the three in one place from points is missing in the literature. Our algorithm estimates the normals at each sample point through poles (farthest Voronoi vertices of a sample point) and uses the estimated normals and the corresponding tangents to determine the spatial locations (inner or outer) of the Voronoi vertices with respect to the original curve. The vertex classification helps to construct a piece‐wise linear approximation to the object boundary. We provide a theoretical analysis of the algorithm for points non‐uniformly (ε‐sampling) sampled from simple, closed, concave and smooth curves. The proposed framework has been thoroughly evaluated for its usefulness using various test data. Results indicate that even sparsely and non‐uniformly sampled curves with outliers or collection of curves are faithfully reconstructed by the proposed algorithm.

Abstract Given a planar point set sampled from an object boundary, the process of approximating the original shape is called curve reconstruction. In this paper, a novel non‐parametric curve reconstruction algorithm based on Delaunay triangulation has been proposed and it has been theoretically proved that the proposed method reconstructs the original curve under ε‐sampling. Starting from an initial Delaunay seed edge, the algorithm proceeds by finding an appropriate neighbouring point and adding an edge between them. Experimental results show that the proposed algorithm is capable of reconstructing curves with different features like sharp corners, outliers, multiple objects, objects with holes, etc. The proposed method also works for open curves. Based on a study by a few users, the paper also discusses an application of the proposed algorithm for reconstructing hand drawn skip stroke sketches, which will be useful in various sketch based interfaces.

Voronoi diagram (VD) is an extensively studied geometric entity since it has applications in fields such as computer graphics, computer vision, geometric modeling etc. In this paper, an algorithm for computing the VD of a set of planar freeform closed convex curves has been developed, without approximating the curves using points or lines. Algorithms for VD predominantly lie on computing the bisectors, a geometrically complex and a high degree curve even for inputs of low degree. Hence, a lot of processing is required to compute bisector segments that contribute to VD as well as in the computation of branch points. In this paper, it has been shown that computation of a branch point is possible without first computing either the bisector or segments of it. The algorithm uses the minimum antipodal discs (MADs) for all pairs of curves. Three touch discs (TTD) (circles touching three curves) are computed only for a specific set of three curves. Decision criteria for a TTD to become a branch disc (BD, i.e. empty TTD) have been addressed without using explicit curve containment check. Local computations of Voronoi segments are then done. The algorithm gives all branch points via centers of the branch discs along with the segments of curves that will contribute to VD. Results of implementation are provided along with analysis of the algorithm.

This paper presents the summary of all the results of the participants in the event SHREC08 — CAD Model Track

We present an algorithm for generating the Voronoi cells for a set of rational C1-continuous planar closed curves, which is precise up to machine precision. Initially, bisectors for pairs of curves, (C(t), Ci(r)), are generated symbolically and represented as implicit forms in the tr-parameter space. Then, the bisectors are properly trimmed after being split into monotone pieces. The trimming procedure uses the orientation of the original curves as well as their curvature fields, resulting in a set of trimmed-bisector segments represented as implicit curves in a parameter space. A lower-envelope algorithm is then used in the parameter space of the curve whose Voronoi cell is sought. The lower envelope represents the exact boundary of the Voronoi cell.

This paper presents an algorithm for generating the medial axis transform (MAT) of 3D objects with free-form boundaries. The algorithm proposed uses the exact representation of the part and generates an approximate rational spline description (to within a defined tolerance) of the MAT. The algorithm generates the MAT by a tracing technique that marches along the object boundary. The level of approximation is controlled by the choice of the step size in the tracing procedure. Criteria based on distance and local curvature of boundary entities are used to identify the junction points and the search for these junction points is done in an efficient way. The algorithm works for multiply-connected objects as well. Results of implementation are provided.