1,310 Hits in 3.2 sec

Multi-domain semantic segmentation with overlapping labels [article]

Petra Bevandić, Marin Oršić, Ivan Grubišić, Josip Šarić, Siniša Šegvić
2021 arXiv   pre-print
We address this challenge by proposing a principled method for seamless learning on datasets with overlapping classes based on partial labels and probabilistic loss.  ...  Our method achieves competitive within-dataset and cross-dataset generalization, as well as ability to learn visual concepts which are not separately labeled in any of the training datasets.  ...  Thus, we review semantic segmentation, efficient architectures, learning with partial labels, and multi-domain dense prediction.  ... 
arXiv:2108.11224v2 fatcat:qqxnvmcckbftbmonvaf4qx5cpq

A Fast Knowledge Distillation Framework for Visual Recognition [article]

Zhiqiang Shen, Eric Xing
2021 arXiv   pre-print
ReLabel, a recently proposed solution, suggests creating a label map for the entire image.  ...  When conducting multi-crop in the same image for data loading, our FKD is even more efficient than the traditional image classification framework.  ...  ) for some outlier regions, our strategy is substantially more robust than ReLabel, such as the loose bounding boxes of objects, partial object, etc., as shown in Fig. 2 (row 3); (iii) In some particular  ... 
arXiv:2112.01528v1 fatcat:jppaeudnfreprkgaob3hiyxgha

mDALU: Multi-Source Domain Adaptation and Label Unification with Partial Datasets [article]

Rui Gong, Dengxin Dai, Yuhua Chen, Wen Li, Luc Van Gool
2021 arXiv   pre-print
This paper formulates this as a multi-source domain adaptation and label unification problem, and proposes a novel method for it.  ...  In the latter, knowledge is transferred in the unified label space after a label completion process with pseudo-labels.  ...  Standard supervised learning is then performed in the unified label space for the final model.  ... 
arXiv:2012.08385v2 fatcat:mov5x43bjjbgljzkcalkhyaxt4

Fairness-Aware Process Mining [article]

Mahnaz Sadat Qafari, Wil van der Aalst
2019 arXiv   pre-print
for delays).  ...  Process mining is a multi-purpose tool enabling organizations to improve their processes.  ...  Note that in all the experiments the relabeling technique with no limitation on the label of the leaf have been used, which means, both leaves with + and -labels may get relabeled.  ... 
arXiv:1908.11451v1 fatcat:xxulxmoldrea5af47jonkcnpou

Deep Mining External Imperfect Data for Chest X-ray Disease Screening [article]

Luyang Luo, Lequan Yu, Hao Chen, Quande Liu, Xi Wang, Jiaqi Xu, Pheng-Ann Heng
2020 arXiv   pre-print
Specifically, the imperfect data is in two folds: domain discrepancy, as the image appearances vary across datasets; and label discrepancy, as different datasets are partially labeled.  ...  To this end, we formulate the multi-label thoracic disease classification problem as weighted independent binary tasks according to the categories.  ...  [28] studied the multi-organ segmentation with partially-labeled domains and developed a prior-aware neural network. To address missing labels in multi-label classification, Yang et al.  ... 
arXiv:2006.03796v1 fatcat:4wbdav3cqvbo5d6dzy3ravnuia

Label-Noise Robust Multi-Domain Image-to-Image Translation [article]

Takuhiro Kaneko, Tatsuya Harada
2019 arXiv   pre-print
Multi-domain image-to-image translation is a problem where the goal is to learn mappings among multiple domains.  ...  To overcome this limitation, we propose a novel model called the label-noise robust image-to-image translation model (RMIT) that can learn a clean label conditional generator even when noisy labeled data  ...  Acknowledgement This work was partially supported by JST CREST Grant Number JPMJCR1403, Japan, and partially supported by the Ministry of Education, Culture, Sports, Science and Technology (MEXT) as "Seminal  ... 
arXiv:1905.02185v1 fatcat:vbyukorjyrhanmfnm5usfilkxy

Monotonic classification: an overview on algorithms, performance measures and data sets [article]

José-Ramón Cano and Pedro Antonio Gutiérrez and Bartosz Krawczyk and Michał Woźniak and Salvador García
2018 arXiv   pre-print
the target class label should not decrease when input attributes values increase).  ...  There are many real-world scenarios, such as bankruptcy prediction, option pricing or medical diagnosis, where the classification models to be learned need to fulfil restrictions of monotonicity (i.e.  ...  The proposed method consists of two separated stages for learning and subsequent tuning.  ... 
arXiv:1811.07155v1 fatcat:h7uqontgl5gdtgr3bduu73wg44

Learning to Segment from Scribbles using Multi-scale Adversarial Attention Gates [article]

Gabriele Valvano, Andrea Leo, Sotirios A. Tsaftaris
2021 arXiv   pre-print
We also demonstrate extensions in a variety of settings: semi-supervised learning; combining multiple scribble sources (a crowdsourcing scenario) and multi-task learning (combining scribble and mask supervision  ...  With unpaired segmentation masks, we train a multi-scale GAN to generate realistic segmentation masks at multiple resolutions, while we use scribbles to learn their correct position in the image.  ...  To improve the learning process and avoid overfitting, we make the adversarial game harder for the discriminator, using label noise [52] and instance noise [53] .  ... 
arXiv:2007.01152v3 fatcat:zoiczzbhmrgq5nqvs3omq3afva

Object Boundary Guided Semantic Segmentation [article]

Qin Huang, Chunyang Xia, Wenchao Zheng, Yuhang Song, Hao Xu, C.-C. Jay Kuo
2016 arXiv   pre-print
relabeled boundaries.  ...  To tackle this problem, we introduce a double branch fully convolutional neural network, which separates the learning of the desirable semantic class labeling with mask-level object proposals guided by  ...  The network learns to design tailored feature pools for a vision task by examining deep features of discriminative properties and shallow features of local visual patterns.  ... 
arXiv:1603.09742v4 fatcat:6vk75tdmjvgkpdqkkco35qejei

Discriminative, Semantic Segmentation of Brain Tissue in MR Images [chapter]

Zhao Yi, Antonio Criminisi, Jamie Shotton, Andrew Blake
2009 Lecture Notes in Computer Science  
It uses discriminative Random Decision Forest classification and takes into account partial volume effects.  ...  This is combined with correction of intensities for the MR bias field, in conjunction with a learned model of spatial context, to achieve accurate voxel-wise classification.  ...  We also showed that further improvements could be obtained in classification error performance by taking account of partial volume effect, and this suggests a modified test paradigm for future studies.  ... 
doi:10.1007/978-3-642-04271-3_68 fatcat:f7ufaksevvedldhtbzlajg2js4

Automatic Noisy Label Correction for Fine-Grained Entity Typing [article]

Weiran Pan, Wei Wei, Feida Zhu
2022 arXiv   pre-print
In this paper, we propose a novel approach to automatically correct noisy labels for FET without external resources.  ...  Specifically, it first identifies the potentially noisy labels by estimating the posterior probability of a label being positive or negative according to the logits output by the model, and then relabel  ...  The authors would also like to thank the anonymous reviewers for their comments on improving the quality of this paper.  ... 
arXiv:2205.03011v2 fatcat:i64inxlgizbrpogomgepvgchie

Graph-shifts: Natural image labeling by dynamic hierarchical computing

Jason J. Corso, Alan Yuille, Zhuowen Tu
2008 2008 IEEE Conference on Computer Vision and Pattern Recognition  
For low-level vision, we explore image restoration, and for high-level vision, we make use of a hybrid discriminative-generative model to segment and label images into semantically meaningful regions (  ...  In this paper, we present a new approach for image labeling based on the recently introduced graph-shifts algorithm.  ...  Acknowledgements ZT and JJC (partially) are funded by NIH Grant U54 RR021813 entitled Center for Computational Biology. AY is funded by NSF 0413214.  ... 
doi:10.1109/cvpr.2008.4587490 dblp:conf/cvpr/CorsoYT08 fatcat:mbrbzit2evdmbh6tm3zgjnhwda

P-N learning: Bootstrapping binary classifiers by structural constraints

Zdenek Kalal, Jiri Matas, Krystian Mikolajczyk
2010 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition  
We propose a novel paradigm for training a binary classifier from labeled and unlabeled examples that we call P-N learning.  ...  The learning process is guided by positive (P) and negative (N) constraints which restrict the labeling of the unlabeled set.  ...  (iii) Figure 3 . 3 P-N learning first trains a classifier from labeled data and then iterates over: (i) labeling of the unlabeled data by the classifier, (ii) identification and relabeling of examples  ... 
doi:10.1109/cvpr.2010.5540231 dblp:conf/cvpr/KalalMM10 fatcat:4tih4u5cx5ezbn3pp6ayorerb4

Integrated Foreground Segmentation and Boundary Matting for Live Videos

Minglun Gong, Yiming Qian, Li Cheng
2015 IEEE Transactions on Image Processing  
The usage of two competing local classifiers, as we have advocated, provides higher discriminative power while allowing better handling of ambiguities.  ...  By exploiting this proposed machine learning technique, and by addressing both foreground segmentation and boundary matting problems in an integrated manner, our algorithm is shown to be particularly competent  ...  The partially trained 1SVMs are thus directly used to perform relabeling.  ... 
doi:10.1109/tip.2015.2401516 pmid:25675459 fatcat:hnnh6vvvwjeshdnfrkhc43pgyq

Data preprocessing techniques for classification without discrimination

Faisal Kamiran, Toon Calders
2011 Knowledge and Information Systems  
discrimination without relabeling instances.  ...  We first study the theoretically optimal trade-off between accuracy and non-discrimination for pure classifiers.  ...  Acknowledgments We thank the anonymous reviewers for their insightful comments and the many suggestions that contributed substantially to the improvement of the document.  ... 
doi:10.1007/s10115-011-0463-8 fatcat:4xytwosddjhnhmri5nlovcmcse
« Previous Showing results 1 — 15 out of 1,310 results