FGCNet: Fast Graph Convolution for Matching Features

Liu Liu

Abstract:

This paper proposes a fast graph convolution network (FGCNet) to match two sets of sparse features. The network has three new
modules connected in sequence: (i) a local graph convolution block takes point-wise features as inputs and encodes local contextual information to extract local features; (ii) a fast graph message-passing network takes local features as inputs, encodes two-view global
contextual information, to improve the discriminativeness of pointwise features; (iii) a preemptive Optimal Transport (OT) layer takes
point-wise features as inputs, regress point-wise matchedness scores and estimate a 2D joint probability matrix, with each item describes
the matchedness of a feature correspondence. We validate the proposed method on three AR/VR related tasks: two-view matching,
3D reconstruction and visual localization. Experiments show that our method reduces computational complexity significantly compared with state-of-the-art methods, while competitive performance is achieved.