The Generalized PatchMatch Correspondence Algorithm

By Connelly Barnes, Eli Shechtman, Dan B. Goldman, Adam Finkelstein. European Conference on Computer Vision, 2010

616 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/41765f1f10bf2830c7749d8672eeabcb028381ae

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

FlowWeb: Joint image set alignment by weaving consistent, pixel-wise correspondences (2015, methodology): “Other works in this space include [2], which generalizes PatchMatch [1] to use feature descriptors instead of pixel patches, and more recently, finding pairwise correspondences using convolutional features [26].”
AutoScaler: Scale-Attention Networks for Visual Correspondence (2016, methodology): “To compute dense correspondences using these descriptors, one needs to impose smoothness prior to regularize the correspondence map from the sparse matches, which often experiences loss in accuracy [48, 15, 6].”
Flow Fields: Dense Correspondence Fields for Highly Accurate Large Displacement Optical Flow Estimation (2015, methodology): “The propagation of our initial ﬂow values works similar to the propagation step in the PatchMatch approach [4] i.”
Optical flow modeling and computation: A survey (2015, methodology): “It is easy to extend the matching to more complex transformations than translation, like rotation or scale factor [16] by increasing the degree of the search space.”
Shape Completion from a Single RGBD Image (2017, methodology): “[27] proposed the Generalized PatchMatch algorithm, which ef ciently nds dense approximate nearest source patches for all target image patches, with rotations and scales of source patches considered.”
Single image super-resolution from transformed self-exemplars (2015, methodology): “In such cases, our algorithm reduces to searching for affine transformations only in the nearest neighbor field, similar to [2].”
Fast Optical Flow Using Dense Inverse Search (2016, methodology): “In the context of scene correspondence, the SIFTflow [33] and PatchMatch [5] algorithms use descriptors or small patches.”
Contextual-Based Image Inpainting: Infer, Match, and Translate (2017, background): “Traditional non-neural inpainting algorithms [2, 3] mostly work on the image space.”
A global sampling method for alpha matting (2011, methodology): “A recent extension of PatchMatch [4] takes into account patch rotation and scaling.”
Shift-Net: Image Inpainting via Deep Feature Rearrangement (2018, background): “ng best matching patches from the known region. For fast patch search, Barnes et al. suggest a PatchMatch algorithm [3] to exploit the image coherency, and generalize it for nding k-nearest neighbors [11]. Generally, exemplar-based inpainting is superior in synthesizing textures, but is not well sui”
Learning to group discrete graphical patterns (2017, neutral): “Many techniques have been developed for exact and approximate symmetry detection of patterns found in natural images and shapes, e.g., [Barnes et al. 2010; Graham et al. 2010; Lukac et al. 2017; Podolak et al. 2006]; see also surveys by [Liu et al. 2010; Mitra et al. 2012].”
Transformation guided image completion (2013, background): “, scale [16], similarity transform [19, 4, 10] and similarity with flip over y axis [7].”
Image Inpainting : Overview and Recent Advances (2014, methodology): “The PatchMatch algorithm has been generalized in [30] to find K -NNs instead of one, and to extend the search space with transformations (rotations, scaling).”
Projection onto the Manifold of Elongated Structures for Accurate Extraction (2015, methodology): “Therefore, since ΠN (X) is defined in (2) as the point of minimum distance to X inMN , we have ‖ΠN (X)−X‖(2) ≤ ‖ΠD→N (X)−X‖(2).”
Learning Mid-level Filters for Person Re-identification (2014, background): “Patch matching is widely used, and many off-the-shelf fast algorithms [1] are available for faster speed.”
Unsupervised Salience Learning for Person Re-identification (2013, neutral): “Many off-theshelf methods [2, 3] are available to boost the performance and efficiency.”
Unsupervised Deep Learning for Optical Flow Estimation (2017, methodology): “Xu et. al (Xu, Jia, and Matsushita 2012) further integrates more advanced matching methods such as SIFT-flow (Liu, Yuen, and Torralba 2011) and PatchMatch (Barnes et al. 2010) to increase the accuracy of flow.”
Internal statistics of a single natural image (2011, result): “A similar observation was also made by [2].”
Person Re-identification by Salience Matching (2013, methodology): “Patch matching is widely used, and many off-theshelf methods [1] are available.”