Pixel-Wise Symbol Spotting via Progressive Points Location for Parsing CAD Images

TL;DR

The paper tackles pixel-level symbol spotting in CAD images by introducing Progressive Gaussian Kernels (PGK) to anneal the Gaussian heatmap size during training and a local offset to reduce quantization errors in coordinate decoding, thereby enabling accurate pixel-wise keypoint localization. It further adds a symbol-grouping mechanism to reconstruct rectangle symbols from keypoints, and demonstrates its approach on a telecommunication equipment-room CAD image dataset, achieving improvements over heatmap-only and regression baselines. The contributions include (i) PGK with an annealing schedule for stable, high-signal training, (ii) a local offset encoding to minimize discretization errors, (iii) a grouping algorithm for rectangle symbols, and (iv) a new CAD image dataset for symbol spotting. Collectively, the method provides a simple yet effective framework for CAD image parsing with low labeling cost and strong real-world generalization, potentially enabling better CAD revision, semantic management, and rapid 3D prototyping.

Abstract

Parsing Computer-Aided Design (CAD) drawings is a fundamental step for CAD revision, semantic-based management, and the generation of 3D prototypes in both the architecture and engineering industries. Labeling symbols from a CAD drawing is a challenging yet notorious task from a practical point of view. In this work, we propose to label and spot symbols from CAD images that are converted from CAD drawings. The advantage of spotting symbols from CAD images lies in the low requirement of labelers and the low-cost annotation. However, pixel-wise spotting symbols from CAD images is challenging work. We propose a pixel-wise point location via Progressive Gaussian Kernels (PGK) to balance between training efficiency and location accuracy. Besides, we introduce a local offset to the heatmap-based point location method. Based on the keypoints detection, we propose a symbol grouping method to redraw the rectangle symbols in CAD images. We have released a dataset containing CAD images of equipment rooms from telecommunication industrial CAD drawings. Extensive experiments on this real-world dataset show that the proposed method has good generalization ability.

Figures (10)

• Figure 1: 12 kinds of semantic objects are defined within CAD images. A red box represents the region symbol, while the green wireframes represent the rectangle symbols.
• Figure 2: Comparison between a CAD drawing and a CAD image.
• Figure 3: Illustration of MVD problem by comprising between the GT heatmaps and the predicted heatmaps. (a) and (b) represent the GT heatmaps with the GKS $\sigma=3$ and $\sigma=1$,respectively. (c) and (d) show the predicted heatmaps with the corresponding GKS respectively. Color represents the magnitude of the heatmaps response. Circle indicates the position with the maximum value.
• Figure 4: The relationship between the gradients \ref{['eqt4:gradient']} and the distance $|x-\mu_x|$.
• Figure 5: The loss curve of the validation set for PGK and naive approach. At 100 epochs, the naive method changes the GKS from 3 to 1.