End-to-End Wireframe Parsing

Abstract

We present a conceptually simple yet effective algorithm to detect wireframes [14] in a given image. Compared to the previous methods [14,33] which first predict an intermediate heat map and then extract straight lines with heuristic algorithms, our method is end-to-end trainable and can directly output a vectorized wireframe that contains semantically meaningful and geometrically salient junctions and lines. To better understand the quality of the outputs, we propose a new metric for wireframe evaluation that penalizes overlapped line segments and incorrect line connectivities. We conduct extensive experiments and show that our method significantly outperforms the previous state-of-the-art wireframe and line extraction algorithms [14,33,32]. We hope our simple approach can be served as a baseline for future wireframe parsing studies. Code has been made publicly available at https://github.com/zhou13/lcnn.

Wireframe Parsing With Guidance of Distance Map

“…For AFM, aspect ratio is used to filter out false detections, and it varies in the range (0, 1] with a step size of 0.1. The Wireframe Parser rejects false detections by applying an array of thresholds [2,6,10,20,30,50,80,100,150,200,250,255] to binarize the line heatmap, while keeping the threshold of junction confidence and junction branch confidence fixed. In the proposed method, we simply vary the threshold of the confidence of line segment detections in the range (0, 1] with a step size of 0.1 to pick true detections.…”

“…Huang et al [6] train a anchor-based junction detector using deep neural networks on a large-scale dataset with junction annotation, and achieve the state-of-the-art results. Recently, Zhou et al [20] propose to find junctions in an end-to-end manner.…”

ChainLineNet: Deep-Learning-Based Segmentation and Parameterization of Chain Lines in Historical Prints

“…To train their method, they created a large wireframe benchmark dataset. Zhou et al [10] designed an end-to-end trainable L-CNN that directly predicts vectorized wireframes. The L-CNN consists of a stacked hourglass network as the feature extraction backbone, a heat-map-based junction proposal module, a line-sampling module that generates line candidates based on the predicted junctions, and a line verification module, for which the line of interest (LoI) pooling layer is utilized, which compares line segments with corresponding positions in the feature maps of the backbone.…”

“…Lin et al [12] proposed in their deep Hough transform line priors method to combine line priors with deep learning by incorporating a trainable Hough transform block into a deep network and performing filtering in the Hough domain with local convolutions. For the application of line detection on the Wireframe datasets, they used the L-CNN [10] and the HAWP [11] as backbones and replaced the hourglass blocks with their Hough transform blocks.…”

LiDAR-Camera Calibration Using Line Correspondences

“…X ∞ and x ∞ make up a 3D-2D infinity point pair. Here, we use L-CNN [40] to detect the edges on an image. A RANSAC procedure is used to detect the infinity points of artificial buildings from images as in [41][42][43].…”