Matching with AffNet based rectifications
This addresses the problem of efficient image matching for computer vision applications, but it is incremental as it builds on existing AffNet and view synthesis techniques.
The paper tackles two-view matching under significant viewpoint changes by proposing two methods, DenseAffNet and DepthAffNet, to minimize view synthesis overhead. DenseAffNet is faster than state-of-the-art and more accurate on generic scenes, while DepthAffNet matches state-of-the-art on scenes with large planes, as evaluated on three public datasets.
We consider the problem of two-view matching under significant viewpoint changes with view synthesis. We propose two novel methods, minimizing the view synthesis overhead. The first one, named DenseAffNet, uses dense affine shapes estimates from AffNet, which allows it to partition the image, rectifying each partition with just a single affine map. The second one, named DepthAffNet, combines information from depth maps and affine shapes estimates to produce different sets of rectifying affine maps for different image partitions. DenseAffNet is faster than the state-of-the-art and more accurate on generic scenes. DepthAffNet is on par with the state of the art on scenes containing large planes. The evaluation is performed on 3 public datasets - EVD Dataset, Strong ViewPoint Changes Dataset and IMC Phototourism Dataset.