Structured Forests for Fast Edge Detection

By Piotr Doll'ar, C. Lawrence Zitnick. 2013

989 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Media Technology

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 Media Technology

Citations in the corpus

What Makes for Effective Detection Proposals? (2015, background): “EdgeBoxesyEC [20] also starts from a coarse sliding window pattern, but builds on object boundary estimates (obtained via structured decision forests [36], [42]) and adds a subsequent refinement step to improve localisation.”
Extreme Clicking for Efficient Object Annotation (2017, methodology): “4, we use the extreme clicks to estimate an initial contour of the object by following predicted object boundaries [15].”
Fast Edge Detection Using Structured Forests (2014, background): “Finally, we note that sweep details have changed slightly from the our previous work [14]; most notably, the sweeps now utilize”
Learning Rich Features from RGB-D Images for Object Detection and Segmentation (2014, methodology): “To make the model generalize better, we add the soft edge map produced by running the RGB edge detector of [9] (which is trained on BSDS) on the RGB image.”
Learning visual groups from co-occurrences in space and time (2015, methodology): “However, rather than using a hand-engineered grouping cue Shi & Malik (2000); Zitnick & Doll´ar (2014), or learning to group with direct supervision Doll´ar & Zitnick (2013); Krahnenbuhl & Koltun (2015), we use a measure of spatial and temporal dependence as the afﬁnity.”
Designing deep networks for surface normal estimation (2014, methodology): “For visualization purposes, we project these edge labels onto the output of Structured Edges [6].”
Picture: A probabilistic programming language for scene perception (2015, methodology): “Representation Layer: Transforms ID or IR into a hierarchy of coarse-to-fine image representations V(ID) and v(h) (deep neural networks [25, 23], contours [8] and pixels).”
DeepEdge: A multi-scale bifurcated deep network for top-down contour detection (2014, methodology): “Shift [8] 0.64 0.68 0.56 Ncuts [26] 0.64 0.68 0.45 Felz., Hutt. [10] 0.61 0.64 0.56 SCT [17] 0.71 0.72 0.74 gPb-owt-ucm [2] 0.73 0.76 0.73 SCG [23] 0.74 0.76 0.77 Sketch Tokens [16] 0.73 0.75 0.78 SE [9] 0.74 0.76 0.78 PMI [12] 0.74 0.77 0.78 DeepEdge 0.75 0.77 0.81 Table 1: Edge detection results o”
Reading Text in the Wild with Convolutional Neural Networks (2014, methodology): “Following [62], we compute the edge response map using the Structured Edge detector [15, 16] and perform Non-Maximal Suppression orthogonal to the edge responses, sparsifying the edge map.”
Deep Layer Aggregation (2017, methodology): “SE [10] 0.”
Multiscale Combinatorial Grouping for Image Segmentation and Object Proposal Generation (2015, methodology): “Single-scale segmentation: We consider as input the following local contour cues: (1) brightness, color and texture differences in half-disks of three sizes [33], (2) sparse coding on patches [22], and (3) structured forest contours [34].”
BDD100K: A Diverse Driving Dataset for Heterogeneous Multitask Learning (2018, methodology): “We evaluate drivable area segmentation by mean IoU, and lane marking by the Optimal Dataset Scale F-measure (ODS-F) for each category of the three attributes using the Structured Edge Detection Toolbox [10] with tolerance τ = 1, 2, and 10 pixels.”
N4-Fields: Neural Network Nearest Neighbor Fields for Image Transforms (2014, methodology): “Committee of N 4 -ﬁelds.”
Crisp Boundary Detection Using Pointwise Mutual Information (2014, methodology): “Speed: Recently several edge detectors have been proposed that optimize speed while also achieving good results [8,7].”
Oriented edge forests for boundary detection (2014, neutral): “For example, [22] learns weightings for each sparse code channel and hypothesized edge orientation while [8, 16] predict the probability of a boundary at an image location using a cascade or randomized decision forest built over simple image features.”