Learning to segment under various forms of weak supervision

By Jia Xu, Alexander Gerhard Schwing, Raquel Urtasun. 2015

190 citations recorded by the source corpus.

Subfields: Artificial Intelligence, 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 Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

The Cityscapes Dataset for Semantic Urban Scene Understanding (2016, neutral): “Last but not least, several studies [5,11,48–50,53,74,76] lately have explored different forms of weak supervision, such as bounding boxes or image-level labels, for training CNNs for pixel-level semantic labeling.”
COCO-Stuff: Thing and Stuff Classes in Context (2016, background): “Is stuff easier than things? Several works found that stuff is easier to segment than things [50, 28, 36, 51, 53, 62, 60, 63].”
Colorization as a Proxy Task for Visual Understanding (2017, background): “A related endeavor to unsupervised learning is developing models that work with weaker forms of supervision [2, 39].”
Weakly-and Semi-Supervised Learning of a Deep Convolutional Network for Semantic Image Segmentation (2015, methodology): “Bounding box annotations have been utilized for semantic segmentation by [38, 42], while [15, 21, 40] describe schemes exploiting both image-level labels and bounding box annotations.”
What's the Point: Semantic Segmentation with Point Supervision (2015, methodology): “[41] also trains weakly-supervised models with squiggles, but they randomly generate these squiggles from the ground truth, whereas we collect them from humans.”
No More Discrimination: Cross City Adaptation of Road Scene Segmenters (2017, background): “Alternatively, [15, 38] apply free-form squiggles to provide partial pixel labels for data collection.”
Pseudo-mask Matters in Weakly-supervised Semantic Segmentation (2021, methodology): “To reduce the annotation burden, weaklysupervised semantic segmentation approaches have been proposed using scribble annotations [23, 32], bounding boxes [36, 9, 16] , points [3] or image-level labels [17, 27, 28, 2, 33].”
Exploiting Saliency for Object Segmentation from Image Level Labels (2017, methodology): “The weakly supervised learning problem can be seen as a specific instance of learning from constraints [38, 47].”
Simple Does It: Weakly Supervised Instance and Semantic Segmentation (2016, background): “In order to keep annotation cost low, recent work has explored different forms of supervision for semantic labelling: image labels [29, 28, 27, 30, 42], points [3], scribbles [44, 23], and bounding boxes [8, 27].”
Normalized Cut Loss for Weakly-Supervised CNN Segmentation (2018, methodology): “However, we take advantage of deep CNN rather than SVM used in [51].”
Recent progress in semantic image segmentation (2018, background): “More details can refer to Papandreou et al. (2015), Xia et al. (2013), Zhu et al. (2014), Xu et al. (2015).”
Weakly-Supervised Semantic Segmentation Network with Deep Seeded Region Growing (2018, methodology): “Various weakly-supervised methods have been proposed for learning to perform semantic segmentation with coarser annotations, such as image labels [20, 36], points [2], scribbles [16], and bounding”
Seed, Expand and Constrain: Three Principles for Weakly-Supervised Image Segmentation (2016, methodology): “Methods in the tradition of self-training [29] train a fullysupervised model but create the necessary pixel-level annotation using the model itself in an EM-like procedure [43,44,48].”
Learning Random-Walk Label Propagation for Weakly-Supervised Semantic Segmentation (2017, methodology): “Our method can be viewed from a similar perspective, since we also impose our weak supervision via linear constraints (7); however, our constraints explicitly model the process of spatial label propagation, whereas the constraints proposed in [15, 13] model only aggregate statistics over regions, an”
Saliency Guided Self-Attention Network for Weakly and Semi-Supervised Semantic Segmentation (2019, background): “This supervision manner requires a large amount of pixel-wise To reduce the annotation burden, different supervision forms such as bounding boxes [8], scribbles [9], and image-level tags [10] have been considered for semantic segmentation. expensive and time-consuming.”
Object Region Mining with Adversarial Erasing: A Simple Classification to Semantic Segmentation Approach (2017, background): “Some works [16–19, 27, 31] propose to train the segmentation models by only using image-level labels, which is the simplest supervision for training semantic segmentation models.”
Spot On: Action Localization from Pointly-Supervised Proposals (2016, background): “Obvious candidates for the supervision are action class labels and bounding boxes, but other forms of supervision are also feasible [11].”
Interactive Full Image Segmentation by Considering All Regions Jointly (2018, background): “Many works address semantic segmentation by training from weakly labeled data, such as image-level labels [28, 44, 58], point-clicks [5, 6, 13, 57], boxes [26, 37, 41] and scribbles [33, 59].”