Semantic contours from inverse detectors

By Bharath Hariharan, Pablo Arbeláez, Lubomir D. Bourdev, Subhransu Maji, Jitendra Malik. 2011

1,892 citations recorded by the source corpus.

We study the challenging problem of localizing and classifying category-specific object contours in real world images. For this purpose, we present a simple yet effective method for combining generic object detectors with bottom-up contours to identify object contours. We also provide a principled way of combining information from different part detectors and across categories. In order to study the problem and evaluate quantitatively our approach, we present a dataset of semantic exterior boundaries on more than 20, 000 object instances belonging to 20 categories, using the images from the VOC2011 PASCAL challenge [7].

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

Fully convolutional networks for semantic segmentation (2014, methodology): “[14] collected labels for a larger set of 8498 PASCAL training images, which was used to train the previous state-of-theart system, SDS [15].”
Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation (2013, background): “Since some classes have very few examples in val (the smallest has only 31 and half have fewer than 110), it is important to produce an approximately class-balanced partition.”
Weight Standardization (2019, methodology): “Following common practice, the training set is augmented by the annotations provided in [9], resulting in 10,582 images.”
Revisiting Unreasonable Effectiveness of Data in Deep Learning Era (2017, methodology): “As is standard practice, all models are trained on an augmented PASCAL VOC [12] 2012 ‘trainaug’ set with 10582 images (extra annotations from [15]).”
Learning Features by Watching Objects Move (2016, methodology): “[18], and we test on the VOC 2011 val set after removing overlapping images from SBD train.”
Context Encoders: Feature Learning by Inpainting (2016, methodology): “3, using the additional training data from [18], and removing overlapping images from the validation set [28].”
Recurrent Neural Networks for Semantic Instance Segmentation (2017, methodology): “We use the additional annotations from [16] to train our models and we evaluate on the original VOC 2012 validation set, composed of 1,449 images.”
Fully Convolutional Instance-Aware Semantic Segmentation (2016, methodology): “Following the protocol in [15, 7, 16, 8], model training is performed on the VOC 2012 train set, and evaluation is performed on the VOC 2012 validation set, with the additional instance mask annotations from [14].”
Fully Convolutional Multi-Class Multiple Instance Learning (2014, methodology): “We train and validate on the VOC 2011 train and val sets then evaluate on the completely held-out VOC 2012 test set.”
Pixelwise Instance Segmentation with a Dynamically Instantiated Network (2017, methodology): “We also evaluate our model on the SBD dataset, which consists of 5623 training and 5732 validation images, as shown in Tab.”
Shape-aware Instance Segmentation (2016, methodology): “To this end, we made use of the 5732 Pascal VOC 2012 validation images with ground-truth from [12], and compare our approach with the state-of-the-art segmentation proposal generation methods according to the criteria of [13, 21].”
Instance-Sensitive Fully Convolutional Networks (2016, methodology): “Following [3,4], we use the segmentation annotations from [27], and train the models on the training set and evaluate on the validation set.”
Instance-Aware Semantic Segmentation via Multi-task Network Cascades (2015, methodology): “We use the segmentation annotations in [12] for training and evaluation, following [13, 7, 14].”
Inside-Outside Net: Detecting Objects in Context with Skip Pooling and Recurrent Neural Networks (2015, background): “Legend: 07+12: 07 trainval + 12 trainval, 07+12+S: 07+12 plus SBD segmentation labels [12], R: include 2x stacked 4-dir IRNN (context features), W: two rounds of box regression and weighted voting [8], D: remove all dropout, SS: SelectiveSearch [40], EB: EdgeBoxes [43], RPN: region proposal net.”
Convolutional feature masking for joint object and stuff segmentation (2014, methodology): “2012 and the additional segmentation annotations from [9] are used for training and evaluation as in [4, 8, 10].”
Simultaneous Detection and Segmentation (2014, methodology): “We use the segmentation annotations from SBD [17] to train and evaluate.”
DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs (2016, methodology): “The dataset is augmented by the extra annotations provided by [85], resulting in 10,582 (trainaug) training images.”
SimpleClick: Interactive Image Segmentation with Simple Vision Transformers (2022, methodology): “• SBD [18]: It contains 8498 training images (20172 instances) and 2857 validation images (6671 instances).”
FocalClick: Towards Practical Interactive Image Segmentation (2022, methodology): “Some early works are trained on SBD [15] which is a relatively small train set.”