LSDNet: Trainable Modification of LSD Algorithm for Real-Time Line Segment Detection
This work addresses the need for fast and accurate line segment detection in real-time applications, though it is incremental as it builds on existing LSD and CNN methods.
The paper tackles the trade-off between accuracy and speed in line segment detection by proposing LSDNet, a hybrid method that integrates a lightweight CNN into the classical LSD algorithm, achieving 214 FPS with 78 Fh accuracy on the Wireframe dataset, and they argue that annotation errors reduce the actual accuracy gap to 0.2 Fh.
As of today, the best accuracy in line segment detection (LSD) is achieved by algorithms based on convolutional neural networks - CNNs. Unfortunately, these methods utilize deep, heavy networks and are slower than traditional model-based detectors. In this paper we build an accurate yet fast CNN- based detector, LSDNet, by incorporating a lightweight CNN into a classical LSD detector. Specifically, we replace the first step of the original LSD algorithm - construction of line segments heatmap and tangent field from raw image gradients - with a lightweight CNN, which is able to calculate more complex and rich features. The second part of the LSD algorithm is used with only minor modifications. Compared with several modern line segment detectors on standard Wireframe dataset, the proposed LSDNet provides the highest speed (among CNN-based detectors) of 214 FPS with a competitive accuracy of 78 Fh . Although the best-reported accuracy is 83 Fh at 33 FPS, we speculate that the observed accuracy gap is caused by errors in annotations and the actual gap is significantly lower. We point out systematic inconsistencies in the annotations of popular line detection benchmarks - Wireframe and York Urban, carefully reannotate a subset of images and show that (i) existing detectors have improved quality on updated annotations without retraining, suggesting that new annotations correlate better with the notion of correct line segment detection; (ii) the gap between accuracies of our detector and others diminishes to negligible 0.2 Fh , with our method being the fastest.