Fully-trainable deep matching

By James Thewlis, Shuai Zheng, Philip H. S. Torr, Andrea Vedaldi. 2016

42 citations recorded by the source corpus.

Deep Matching (DM) is a popular high-quality method for quasi-dense image matching. Despite its name, however, the original DM formulation does not yield a deep neural network that can be trained end-to-end via backpropagation. In this paper, we remove this limitation by rewriting the complete DM algorithm as a convolutional neural network. This results in a novel deep architecture for image matching that involves a number of new layer types and that, similar to recent networks for image segmentation, has a U-topology. We demonstrate the utility of the approach by improving the performance of DM by learning it end-to-end on an image matching task.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

FlowNet 2.0: Evolution of Optical Flow Estimation with Deep Networks (2016, background): “[30] formulate Deep Matching [32] as a convolutional network and optimize it end-to-end.”
Unsupervised Learning of Object Landmarks by Factorized Spatial Embeddings (2017, background): “Matching images up to a motion-induced deformation links back to the work of Horn and Schunck [26] on optical flow and to deep learning approaches for its computation [21, 57, 28].”
SfM-Net: Learning of Structure and Motion from Video (2017, background): “Recent works [7, 20, 29] propose learning frame-to-frame motion ﬁelds with deep neural networks supervised with ground-truth motion obtained from simulation or synthetic movies.”
Unsupervised object learning from dense equivariant image labelling (2017, methodology): “The state-of-the-art methods [10, 29, 15] typically employ convolutional neural networks to learn pairwise dense correspondences between the same”
DSAC — Differentiable RANSAC for Camera Localization (2016, methodology): “propagating gradients back through an entire pipeline to allow the direct optimization of a task-specific loss function, examples include [41, 1, 38].”
Convolutional Neural Network Architecture for Geometric Matching (2017, background): “Related are also network architectures for estimating inter-frame motion in video [18, 50, 52] or instance-level homography estimation [14], however their goal is very different from ours, targeting high-precision correspondence with very limited appearance variation and background clutter.”