Filters








2,152 Hits in 4.2 sec

Learning from Pixel-Level Noisy Label : A New Perspective for Light Field Saliency Detection [article]

Mingtao Feng, Kendong Liu, Liang Zhang, Hongshan Yu, Yaonan Wang, Ajmal Mian
2022 arXiv   pre-print
In this paper, we propose to learn light field saliency from pixel-level noisy labels obtained from unsupervised hand crafted featured based saliency methods.  ...  Given this goal, a natural question is: can we efficiently incorporate the relationships among light field cues while identifying clean labels in a unified framework?  ...  Unlike [28, 48, 55] , [49, 54] deals with learning from a single noisy labelling in a much more efficient way. [49] learns saliency prediction and robust fitting models to identify inliers. [54] proposes  ... 
arXiv:2204.13456v1 fatcat:np3aj7mzingu3gt46zwym5emha

Activation to Saliency: Forming High-Quality Labels for Completely Unsupervised Salient Object Detection [article]

Huajun Zhou and Peijia Chen and Lingxiao Yang and Jianhuang Lai and Xiaohua Xie
2021 arXiv   pre-print
In the second stage, a self-rectification learning paradigm strategy is developed to train a saliency detector and refine the pseudo labels online.  ...  In order to overcome these shortcomings, we propose a new two-stage Activation-to-Saliency (A2S) framework that effectively excavates high-quality saliency cues to train a robust saliency detector.  ...  Moreover, instead of extracting noisy saliency cues using traditional SOD methods, we present a novel perspective to excavate high-quality saliency cues based on the learned features of a pre-trained network  ... 
arXiv:2112.03650v3 fatcat:u2lquovonvhhbn2o3lhmqsf424

Model-agnostic Approaches to Handling Noisy Labels When Training Sound Event Classifiers [article]

Eduardo Fonseca, Frederic Font, Xavier Serra
2019 arXiv   pre-print
In this work, we evaluate simple and efficient model-agnostic approaches to handling noisy labels when training sound event classifiers, namely label smoothing regularization, mixup and noise-robust loss  ...  While learning from noisy labels has been an active area of research in computer vision, it has received little attention in sound event classification.  ...  In [16] , two networks operating on different views of the data co-teach each other to learn from noisy labels.  ... 
arXiv:1910.12004v1 fatcat:rpu72uaxmzgn3m24taonrlx6py

Multi-Label Learning from Single Positive Labels

Elijah Cole, Oisin Mac Aodha, Titouan Lorieul, Pietro Perona, Dan Morris, Nebojsa Jojic
2021 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)  
We explore this special case of learning from missing labels across four different multi-label image classification datasets for both linear classifiers and end-to-end finetuned deep networks.  ...  Predicting all applicable labels for a given image is known as multi-label classification.  ...  Since it is easier to train image classifiers on informative labels than uninformative ones, we hypothesize that correct labels are a "good choice" from the algorithm's perspective.  ... 
doi:10.1109/cvpr46437.2021.00099 fatcat:yupyc7xi5rao3ac5qiixmlr2oq

Multi-Label Learning from Single Positive Labels [article]

Elijah Cole, Oisin Mac Aodha, Titouan Lorieul, Pietro Perona, Dan Morris, Nebojsa Jojic
2021 arXiv   pre-print
We explore this special case of learning from missing labels across four different multi-label image classification datasets for both linear classifiers and end-to-end fine-tuned deep networks.  ...  Predicting all applicable labels for a given image is known as multi-label classification.  ...  In the multi-class setting, [26] proposes to learn from complementary labels i.e. they assume access to a single negative label per item that specifies that the item does not belong to a given class.  ... 
arXiv:2106.09708v2 fatcat:xao37rffsbcepdrvtsh4yiidzm

Interactive Label Cleaning with Example-based Explanations [article]

Stefano Teso, Andrea Bontempelli, Fausto Giunchiglia, Andrea Passerini
2021 arXiv   pre-print
We tackle sequential learning under label noise in applications where a human supervisor can be queried to relabel suspicious examples.  ...  Whenever it detects a suspicious example, Cincer identifies a counter-example in the training set that -- according to the model -- is maximally incompatible with the suspicious example, and asks the annotator  ...  Typical strategies to learning from noisy labels include discarding or downweighting suspicious examples and employing models robust to noise [33, 1, 34, 2] , often requiring a non-trivial noise ratio  ... 
arXiv:2106.03922v3 fatcat:jdi7yfmyifcdpglgdb4v375vmu

Label Cleaning Multiple Instance Learning: Refining Coarse Annotations on Single Whole-Slide Images [article]

Zhenzhen Wang, Aleksander S. Popel, Jeremias Sulam
2021 arXiv   pre-print
the state-of-the-art alternatives, even while learning from a single slide.  ...  We present a method, named Label Cleaning Multiple Instance Learning (LC-MIL), to refine coarse annotations on a single WSI without the need of external training data.  ...  Adam Charles, Haoyang Mi and Jacopo Teneggi from the Department of Biomedical Engineering, Johns Hopkins University, for their useful advice and discussions.  ... 
arXiv:2109.10778v1 fatcat:dzf2maywbza63hel7v2okd57wi

Debugging Frame Semantic Role Labeling [article]

Alexandre Kabbach
2019 arXiv   pre-print
We propose a quantitative and qualitative analysis of the performances of statistical models for frame semantic structure extraction.  ...  We report on the robustness of a recent statistical classifier for frame semantic parsing to lexical configurations of predicate-argument structures, relying on an artificially augmented dataset generated  ...  They then used a conditional log-linear model over spans for each role of each evoked where the kappa statistic is used, FrameNet annotators do not have to choose among a fixed pool of label for each annotated  ... 
arXiv:1901.07475v1 fatcat:7sktiu6yjnawnomsqvh5sgde7u

Hybrid Variability Aware Network (HVANet): A Self-Supervised Deep Framework for Label-Free SAR Image Change Detection

Jian Wang, Yinghua Wang, Hongwei Liu
2022 Remote Sensing  
unsupervised label-free SAR image CD by taking inspiration from recent developments in deep self-supervised learning.  ...  In this paper, we argue that these internal hybrid variabilities can also be used for learning stronger feature representation, and we propose a hybrid variability aware network (HVANet) for completely  ...  samples is overly simple, and ii) noisy labels are inevitable.  ... 
doi:10.3390/rs14030734 fatcat:z6gojgx3lnbire6rqcnd32oh44

DeepUSPS: Deep Robust Unsupervised Saliency Prediction With Self-Supervision [article]

Duc Tam Nguyen, Maximilian Dax, Chaithanya Kumar Mummadi, Thi Phuong Nhung Ngo, Thi Hoai Phuong Nguyen, Zhongyu Lou, Thomas Brox
2021 arXiv   pre-print
In this work, we propose a two-stage mechanism for robust unsupervised object saliency prediction, where the first stage involves refinement of the noisy pseudo labels generated from different handcrafted  ...  Each handcrafted method is substituted by a deep network that learns to generate the pseudo labels.  ...  From the robust learning perspective, ? proposes a robust way to learn from wrongly annotated datasets for classification tasks.  ... 
arXiv:1909.13055v4 fatcat:wdx3l53gczcsnbdk4t6ed366ae

Cross-Modal Ranking with Soft Consistency and Noisy Labels for Robust RGB-T Tracking [chapter]

Chenglong Li, Chengli Zhu, Yan Huang, Jin Tang, Liang Wang
2018 Lecture Notes in Computer Science  
Moreover, we propose a single unified optimization algorithm to solve the proposed model with stable and efficient convergence behavior.  ...  Second, we propose an optimal query learning method to handle label noises of queries.  ...  Different from these works, we propose a novel cross-modal ranking algorithm for RGB-T tracking from a new perspective. In particular, our approach has the following advantages. i) Generality.  ... 
doi:10.1007/978-3-030-01261-8_49 fatcat:fccrpevqbzcc3di2btafnvxyxy

Dense Semantic Labeling of Subdecimeter Resolution Images With Convolutional Neural Networks

Michele Volpi, Devis Tuia
2017 IEEE Transactions on Geoscience and Remote Sensing  
Semantic labeling (or pixel-level land-cover classification) in ultra-high resolution imagery (< 10cm) requires statistical models able to learn high level concepts from spatial data, with large appearance  ...  The proposed full patch labeling CNN outperforms these models by a large margin, also showing a very appealing inference time.  ...  ACKNOWLEDGMENTS This work was supported in part by the Swiss National Science Foundation, via the grant 150593 "Multimodal machine learning for remote sensing information fusion" (http://p3.snf.ch/project  ... 
doi:10.1109/tgrs.2016.2616585 fatcat:vkoklqtwjvcqdl7sbfnunaadpm

Predicting Emotion Labels for Chinese Microblog Texts [chapter]

Zheng Yuan, Matthew Purver
2015 Studies in Computational Intelligence  
work described in this paper has been supported by the NSFC Overseas, Hong Kong & Macao Scholars Collaborated Researching Fund (61028003) Acknowledgements The authors would like to thank Ivano Azzini, from  ...  use machine learning techniques to establish a model from a large corpus of reviews.  ...  Machine learning via supervised classification, on the other hand, is robust to such variety but usually requires hand-labeled training data.  ... 
doi:10.1007/978-3-319-18458-6_7 fatcat:orzv7zzxhnewrgrilynvmj27wm

Unsupervised Cell Segmentation and Labelling in Neural Tissue Images

Sara Iglesias-Rey, Felipe Antunes-Santos, Cathleen Hagemann, David Gómez-Cabrero, Humberto Bustince, Rickie Patani, Andrea Serio, Bernard De Baets, Carlos Lopez-Molina
2021 Applied Sciences  
Although various methods have been developed to automate this task, they tend to make use of single-purpose machine learning models that require extensive training, imposing a significant workload on the  ...  A very prominent and useful technique adopted across many different fields is imaging and the analysis of histopathological and fluorescent label tissue samples.  ...  Transfer learning allows storing some learned knowledge from a specific task or data and then applying it to new scenarios [21] .  ... 
doi:10.3390/app11093733 fatcat:ct5d3wtisveytpdvzdgkpo75ne

Detection and Localization of Anomalous Motion in Video Sequences from Local Histograms of Labeled Affine Flows

Juan-Manuel Pérez-Rúa, Antoine Basset, Patrick Bouthemy
2017 Frontiers in ICT  
This work was partially supported by Région Bretagne (Brittany Council) through a contribution to AB's PhD student grant.  ...  The presence of anomalous motion can be detected by deciding that the given motion cannot be fit in a model, which is learned from a set of training data of normal behaviors for a given scenario, computed  ...  Let us also stress that from the perspective of the camera, the cyclist looks not that different from a normal pedestrian.  ... 
doi:10.3389/fict.2017.00010 fatcat:l3g2dllbybhfpia7n7emhsauvi
« Previous Showing results 1 — 15 out of 2,152 results