NeurVPS: Neural Vanishing Point Scanning via Conic Convolution

Yichao Zhou, Haozhi Qi, Jingwei Huang, Yi Ma

Abstract: We present a simple yet effective end-to-end trainable deep network with geometry-inspired new convolution operators for detecting vanishing points in images. Traditional convolution neural networks rely on aggregating local edge/line information and do not have mechanisms to directly exploit the global geometric properties of vanishing points as the intersections of parallel lines in the scene. In this work, we identify a canonical conic space in which global geometric information of vanishing points can be effectively represented and computed locally, and we propose a novel operator named conic convolution which can be implemented as regular convolutions in this space. The new operator explicitly enforces feature extractions and aggregations along the structural lines and yet has the same number of parameters as the regular 2D convolution. Our extensive experiments on both synthetic and real-world datasets show that the proposed operator significantly improves the performance of vanishing point detection over traditional methods.

Code Link: https://github.com/zhou13/neurvps

CMT Num: 471