FlowWeb: Joint image set alignment by weaving consistent, pixel-wise correspondences

By Tinghui Zhou, Yong Jae Lee, Stella X. Yu, Alexei A. Efros. 2015

166 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging

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

In Radiology, Nuclear Medicine and Imaging

Citations in the corpus

Universal Correspondence Network (2016, methodology): “We independently train on the PASCAL dataset [9] with various annotations [4, 36] and on the CUB dataset [30], with the same network architecture.”
Articulation-Aware Canonical Surface Mapping (2020, methodology): “Therefore, we can follow the evaluation protocol typically used for measuring image to image correspondence quality [40, 41], and indirectly evaluate the learned mappings by measuring the accuracy for transferring annotated keypoints from a source to a target image as shown in Figure 7.”
Convolutional Neural Network Architecture for Geometric Matching (2017, methodology): “ences between images is one of the fundamental problems in computer vision[15,21] with applications ranging from large-scale 3D reconstruction [1] to image manipulation [17] and semantic segmentation [45]. Traditionally, correspondences consistent with a geometric model such as epipolar geometry or ”
Weakly Supervised Manifold Learning for Dense Semantic Object Correspondence (2017, methodology): “FlowWeb[34] uses compositions of flow fields for modeling a common image structure consistently among a large set of images of the same category.”
CrDoCo: Pixel-Level Domain Transfer With Cross-Domain Consistency (2019, background): “In semantic matching, cycle or transitivity based consistency loss help regularize the network training [50, 49, 3].”
Recycle-GAN: Unsupervised Video Retargeting (2018, background): “Similar ideas have also found application in learning depth cues in an unsupervised manner [14], machine translation [47], shape correspondences [20], point-wise correspondences [51,52], or domain adaptation [18].”
Unsupervised Visual Representation Learning by Graph-Based Consistent Constraints (2016, methodology): “Cycle consistency in a graph has been applied to various computer vision and graphics problems, including co-segmentation [28,29], structure from motion [30,31] and image matching [32,33].”
Pose Induction for Novel Object Categories (2015, background): “However, the task of discovering correspondences across instances that differ in both pose and appearance, is a particularly challenging one and has been demonstrated only in limited pose and appearance variability [40, 32].”
Dense Semantic Correspondence Where Every Pixel is a Classifier (2015, background): “This allows us to estimate both the likelihood of matches for each pixel individually, and also a global belief of match quality.”
GAN-Supervised Dense Visual Alignment (2021, methodology): “FlowWeb [89] uses cycle consistency constraints to estimate a fully-connected correspondence flow graph.”
Temporal Cycle-Consistency Learning (2019, methodology): “It has been applied successfully for tasks like co-segmentation [43, 44], structure from motion [49, 51], and image matching [54, 55, 56].”
DiscoBox: Weakly Supervised Instance Segmentation and Semantic Correspondence from Box Supervision (2021, background): “For instance, the main semantic correspondence benchmarks [19,39–43] have been focusing on object-centric scenarios which de-emphasizes the role of object localization, while the latest instance segmentation methods do not make use of intra-class correspondences.”
Transfer of View-manifold Learning to Similarity Perception of Novel Objects (2017, background): “There has been a recent trend in exploring inherent information as supervisory signal, including using cycle consistency for learning dense correspondance(Zhou et al., 2015), camera motion for foreground segmentation(Zeng et al., 2016) and context information(Doersch et al., 2015).”
Neural best-buddies (2018, methodology): “ SIFT flow, a number of other flow-based methods were suggested, e.g., Deep flow [Weinzaepfel et al. 2013] and Daisy flow [Yang et al. 2014], which perform matching of visually different scenes, and [Zhou et al. 2015] which proposes a net of correspondences between multiple same-class images, based ”
WarpNet: Weakly Supervised Matching for Single-View Reconstruction (2016, background): “Recently, [34] proposes an unsupervised approach for dense alignment of image sets.”
Learning Dense Correspondence via 3D-Guided Cycle Consistency (2016, methodology): “Our approach could also be thought of as an extension and a reformulation of FlowWeb [40] as a learning problem, where the image collection is stored implicitly in the network representation.”
Unsupervised Learning of Object Landmarks by Factorized Spatial Embeddings (2017, background): “Flow can also be defined semantically rather than geometrically [40, 32, 46, 77, 76].”
Self-Supervised Learning of Geometrically Stable Features Through Probabilistic Introspection (2018, methodology): “Yet, without any further training, they have been shown not to improve over hand-engineered features for geometric tasks [21, 8] and most approaches still combine hand-crafted features and spatial regularization [3, 13, 17, 19, 43].”