Object Contour Detection with a Fully Convolutional Encoder-Decoder Network

By Jimei Yang, Brian L. Price, Scott D. Cohen, Honglak Lee, Ming-Hsuan Yang. 2016

389 citations recorded by the source corpus.

We develop a deep learning algorithm for contour detection with a fully convolutional encoder-decoder network. Different from previous low-level edge detection, our algorithm focuses on detecting higher-level object contours. Our network is trained end-to-end on PASCAL VOC with refined ground truth from inaccurate polygon annotations, yielding much higher precision in object contour detection than previous methods. We find that the learned model generalizes well to unseen object classes from the same supercategories on MS COCO and can match state-of-the-art edge detection on BSDS500 with fine-tuning. By combining with the multiscale combinatorial grouping algorithm, our method can generate high-quality segmented object proposals, which significantly advance the state-of-the-art on PASCAL VOC (improving average recall from 0.62 to 0.67) with a relatively small amount of candidates (~1660 per image).

Subfields: 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 Computer Vision and Pattern Recognition

Citations in the corpus

EDTER: Edge Detection with Transformer (2022, methodology): “DeepEdge [3] CVPR’15 0.753 0.772 0.807 CSCNN [24] ArXiv’15 0.756 0.775 0.798 MSC [51] PAMI’15 0.756 0.776 0.787 DeepContour [48] CVPR’15 0.757 0.776 0.800 HFL [4] ICCV’15 0.767 0.788 0.795 HED [65] ICCV’15 0.788 0.808 0.840 Deep Boundary [29] ICLR’15 0.813 0.831 0.866 CEDN [67] CVPR’16 0.788 0.804 -”
Deep Layer Aggregation (2017, background): “Following [43], we give the boundary labels 10 times weight of the others.”
Recurrent Pixel Embedding for Instance Grouping (2017, methodology): “Quantitatively, we compare our model to COB [64], HED [90], CEDN [91], LEP [66], UCM [1], ISCRA [75], NCuts [78], EGB [24], and the original mean shift (MShift) segmentation algorithm [15].”
One-Shot Video Object Segmentation (2016, methodology): “Offline training: The base CNN of our architecture [47] is pre-trained on ImageNet for image labeling, which has proven to be a very good initialization to other tasks [27, 53, 22, 29, 18, 54].”
Importance-Aware Semantic Segmentation for Autonomous Driving System (2017, result): “SegNet [Vijay Badrinarayanan and Cipolla, 2017] and some other works like [Noh et al., 2015; Hong et al., 2015; Yang et al., 2016] belong to this type.”
Closing the Loop for Robotic Grasping: A Real-time, Generative Grasp Synthesis Approach (2018, methodology): “Fully convolutional networks have been shown to perform well at computer vision tasks requiring transfer between image domains, such image segmentation [1, 19] and contour detection [34].”
Classification With an Edge: Improving Semantic Image Segmentation with Boundary Detection (2016, methodology): “Yang et al. (2016) employ a similar strategy for the opposite purpose: their primary goal is not semantic labeling but rather a “semantically informed” edge detection which accentuates edges that lie on object contours.”
Learning Deep Structured Multi-Scale Features using Attention-Gated CRFs for Contour Prediction (2017, methodology): “Following [38, 41], we use all the training and validation images for learning the proposed model and perform data augmentation as described in [38].”
Deep Structural Contour Detection (2020, methodology): “ODS OIS FPS Canny [5] .611 .676 28 gPb-UCM [2] .729 .755 1/240 SE [10] .743 .763 2.5 DeepContour [36] .757 .776 1 / 30 † DeepEdge [3] .753 .772 1 / 1000 † HFL [4] .767 .788 5 / 6 † HED [46] .788 .808 30 † CEDN [47] .788 .804 10 MIL+G-DSN+MS+NCuts [19] .813 .831 1 RCF-VOC-aug [27] .806 .823 30 † RCF-”
DeepCrack: A deep hierarchical feature learning architecture for crack segmentation (2019, background): “Such application is a task that requires both high-level features and low-level cues [40] .”
Salient Object Detection With Pyramid Attention and Salient Edges (2019, background): “In [23], they considered semantic contour information from a pretrained contour detector [51].”
A Survey on Differential Privacy for Unstructured Data Content (2022, background): “Precision and recall accuracies also evaluate the proportions of original objects among all the detected objects as well as the proportions of detected original objects among all original objects by using state-of-the-art contour detection algorithms [118].”
Self-Supervised Visual Representation Learning from Hierarchical Grouping (2020, background): “Though deep neural networks are now the dominant tool for contour detection [41, 4, 46, 25, 47], the best prior approaches [2, 3, 11] deliver somewhat respectable results.”
Instance-Level Salient Object Segmentation (2017, methodology): “Fully convolutional networks have been widely used in image labeling problems and have achieved great successes in salient region detection [30, 33, 45] and object contour detection [47, 50].”
Video Object Segmentation without Temporal Information (2017, background): “Deconvolutional layers that learn how to upsample are used in [47], [74] to”
Learning robust, real-time, reactive robotic grasping (2019, background): “Fully convolutional networks have been shown to perform well at grasp affordance detection (Zeng et al., 2018a,b), computer vision tasks requiring transfer between image domains, such as image segmentation (Badrinarayanan et al., 2015; Long et al., 2015) and contour detection (Yang et al., 2016).”