Image segmentation with a bounding box prior

By Victor S. Lempitsky, Pushmeet Kohli, Carsten Rother, Toby Sharp. 2009

446 citations recorded by the source corpus.

Subfields: Aerospace Engineering, 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 Aerospace Engineering

In Computer Vision and Pattern Recognition

Citations in the corpus

FocalClick: Towards Practical Interactive Image Segmentation (2022, background): “It enables users to annotate masks conveniently using simple interactions like scribbles [1, 11, 21], boxes [20, 34, 40], or clicks [5, 18, 27, 35, 36, 44].”
Getting to 99% Accuracy in Interactive Segmentation (2020, methodology): “0005 Table 7: Mean IoU for the 1-20 clicks range on the different datasets [15, 16, 37] for each variant of our algorithm.”
Deep Extreme Cut: From Extreme Points to Object Segmentation (2017, background): “One of the most common ways to perform weakly supervised segmentation is drawing a bounding box around the object of interest [4, 38, 32, 18].”
Extreme Clicking for Efficient Object Annotation (2017, methodology): “To reduce annotation time many interactive segmentation techniques have been proposed, which require the user to input either a bounding box around the object [42, 51, 71], or scribbles [3, 16, 24, 26, 27, 42, 50, 64, 65, 72], or clicks [31, 69].”
Weakly-and Semi-Supervised Learning of a Deep Convolutional Network for Semantic Image Segmentation (2015, methodology): “Bounding box annotations are also commonly used in interactive segmentation [22, 33]; we show that such foreground/background”
What's the Point: Semantic Segmentation with Point Supervision (2015, methodology): “[21]. A typical indoor scene contains 23 objects [12], raising the annotation cost to tens of minutes per image. Methods have been developed to reduce the annotation cost through effective interfaces [28, 20, 3, 21, 39], e.g., through requesting human feedback only as necessary [17]. Co-segmentation”
Guide Me: Interacting with Deep Networks (2018, background): “Prior to deep learning, several systems have been proposed for semi-automatic segmentation, that allowed the user to interfere with the result or to provide hints to the system via seed points [8, 52], bounding boxes [24, 37, 54], contours or scribbles [7, 23, 58], eye gaze [26] or in the form of bi”
Y Qualification Main task Y Return N N Y User qualified ? N Feedback Pass ? Pass ? Annotator training Annotating images (2017, methodology): “To reduce annotation time many interactive segmentation techniques have been proposed, which require the user to input either a bounding box around the object [42, 51, 71], or scribbles [3, 16, 24, 26, 27, 42, 50, 64, 65, 72], or clicks [31, 69].”
Harmony Potentials (2011, methodology): “Several approaches are built upon the bounding boxes obtained from a detection method (Lempitsky et al, 2009; Gould et al, 2009; Ladicky et al, 2010b).”
DenseCut: Densely Connected CRFs for Realtime GrabCut (2015, methodology): “In order to achieve high quality results, recent methods have focused on complex formulations [VKR09, LKRS09, TGVB13], which typically lead to slow techniques.”
Geodesic graph cut for interactive image segmentation (2010, background): “This error was also lower than that of any of the other compared-to methods that do not have a spatial position bias, either explicitly [16] or implicitly [12].”
Interactive Image Segmentation via Backpropagating Refinement Scheme (2019, background): “A target object is annotated by a user in the type of bounding box [51, 24, 42] or scribble [52, 11, 10, 25].”
Weakly- and Semi-Supervised Learning of a DCNN for Semantic Image Segmentation (2015, methodology): “ing image segmentation modelsXia et al.(2013);Guillaumin et al.(2014);Chen et al.(2014a);Zhu et al.(2014). Bounding box annotations are also commonly used in foreground/background image segmentation (Lempitsky et al., 2009;Rother et al.,2004). Combining few strong annotations and a large number of w”
Fine-grained recognition without part annotations (2015, methodology): “As noted by [31], objects in bounding boxes typically occupy a non-negligible portion of the bounding box, and we can use this knowledge to significantly improve segmentation quality.”
A survey of graph theoretical approaches to image segmentation (2013, methodology): “[83] used such a rectangle to impose the topological prior on the segmented object.”
Towards Unified Object Detection and Semantic Segmentation (2014, unknown): “In addition, some ambiguous object hypotheses may be difficult to classify from the information within the window/segment alone, but contextual information, such as local context around each object hypothesis and global image-level context, can help [28,34,11].”
Beat the MTurkers: Automatic Image Labeling from Weak 3D Supervision (2014, background): “[20] encodes tightness of the segmentation to the bounding boxes provided by users.”
Exploring Compositional High Order Pattern Potentials for Structured Output Learning (2013, background): “convexity [5], connectivity [32, 23], and a bounding box occupancy constraint [17].”