Filters








14,095 Hits in 7.4 sec

Domain Adaptation for Rare Classes Augmented with Synthetic Samples [article]

Tuhin Das, Robert-Jan Bruintjes, Attila Lengyel, Jan van Gemert, Sara Beery
2021 arXiv   pre-print
As a testbed, we use a camera trap animal dataset with a rare deer class, which is augmented with synthetic deer samples.  ...  While domain adaptation is generally applied on completely synthetic source domains and real target domains, we explore how domain adaptation can be applied when only a single rare class is augmented with  ...  Acknowledgements Computational resources were provided by Microsoft AI for Earth.  ... 
arXiv:2110.12216v1 fatcat:yob7ancqq5dk3av3oqkxuob4qi

Integrating Expert Knowledge with Domain Adaptation for Unsupervised Fault Diagnosis [article]

Qin Wang, Cees Taal, Olga Fink
2021 arXiv   pre-print
In this paper, we aim to overcome this limitation by integrating expert knowledge with domain adaptation in a synthetic-to-real framework for unsupervised fault diagnosis.  ...  dataset is generated by augmenting real vibration samples of healthy bearings.  ...  A synthetic-to-real CWRU experiment with different levels of imbalance on the rolling element fault class. 1% on the x-axis means that only 1% samples are available for rare fault class, compared to the  ... 
arXiv:2107.01849v1 fatcat:hrz4pn73sfbvzpcy4ih5xr6n5m

Image-to-Image Translation of Synthetic Samples for Rare Classes [article]

Edoardo Lanzini, Sara Beery
2021 arXiv   pre-print
One potential approach to increase the training data for these rare classes is to augment the limited real data with synthetic samples.  ...  Compared against a system augmented with unaligned synthetic data, our experiments show a considerable decrease in classification error rates on a rare species.  ...  Acknowledgements We would like to thank the USGS and NPS for providing data and Microsoft AI for Earth for providing compute resources.  ... 
arXiv:2106.12212v1 fatcat:logankwygvcurkfh6eeybyatoq

Traffic Context Aware Data Augmentation for Rare Object Detection in Autonomous Driving [article]

Naifan Li, Fan Song, Ying Zhang, Pengpeng Liang, Erkang Cheng
2022 arXiv   pre-print
We also present a thorough study to illustrate the effectiveness of our local-adaptive and global constraints based Copy-Paste data augmentation for rare object detection.  ...  Specifically, local adaptive instance-level image transformation is introduced to generate realistic rare object masks from source domain to the target domain.  ...  For example, it assigns large weights for tail classes while assigns small weights for head classes.  ... 
arXiv:2205.00376v1 fatcat:byp6oslpwjb3vdfqde64zimmiy

Attention-based Adversarial Appearance Learning of Augmented Pedestrians [article]

Kevin Strauss, Artem Savkin, Federico Tombari
2021 arXiv   pre-print
In this work, we propose a method that leverages the advantages of the augmentation process and adversarial training to synthesize realistic data for the pedestrian recognition task.  ...  Our approach utilizes an attention mechanism driven by an adversarial loss to learn domain discrepancies and improve sim2real adaptation.  ...  Figure 7 . 7 Sample images from No Pedestrians (top) together with their Augmented (middle) and Adapted (bottom) counterparts.  ... 
arXiv:2107.02673v1 fatcat:4bmzrzssyreivcdkslmgjbnmt4

Self-training Improves Pre-training for Natural Language Understanding [article]

Jingfei Du, Edouard Grave, Beliz Gunel, Vishrav Chaudhary, Onur Celebi, Michael Auli, Ves Stoyanov, Alexis Conneau
2020 arXiv   pre-print
Our augmentation approach leads to scalable and effective self-training with improvements of up to 2.6% on standard text classification benchmarks.  ...  To obtain additional data for a specific task, we introduce SentAugment, a data augmentation method which computes task-specific query embeddings from labeled data to retrieve sentences from a bank of  ...  Select top K samples from unlabeled data for each category based on the teacher's prediction to form the synthetically annotated dataset Step 2: Retrieval-based augmentation of in-domain unannotated data  ... 
arXiv:2010.02194v1 fatcat:i4btr6525zb7pe3dd2bfjum3ra

Detection and Localisation of Multiple In-Core Perturbations with Neutron Noise-Based Self-Supervised Domain Adaptation

Aiden Durrant, Georgios Leontidis, Stefanos Kollias, Luis Torres, Cristina Montalvo, Antonios Mylonakis, Christophe Demazière, Paolo Vinai
2021 Zenodo  
-Self-Supervised Domain Adaptation -Synthetic to Real Adaptation Example Prediction Masks Per Class Voxel Prediction Accuracies * No. No.  ...  Unsupervised Domain Adaptation • We aim to learn a discriminative classifier (our voxel-wise semantic segmentation network) for classifying perturbations that is invariant to the presence of a domain shift  ... 
doi:10.5281/zenodo.5575851 fatcat:nd45vlpcsnawzopfyn3hz44zji

Closing the Reality Gap with Unsupervised Sim-to-Real Image Translation [article]

Jan Blumenkamp, Andreas Baude, Tim Laue
2021 arXiv   pre-print
In this paper, we introduce a novel method for augmenting synthetic image data through unsupervised image-to-image translation by applying the style of real world images to simulated images with open source  ...  The generated dataset is combined with conventional augmentation methods and is then applied to a neural network model running in real-time on autonomous soccer robots.  ...  In contrast to end-to-end solutions such as [3, 2] , which integrate the image augmentation into the model, our approach allows to generate a versatile and high quality dataset, which we share with the  ... 
arXiv:1911.01529v2 fatcat:pc5s3d5lhjddbbjvll7bjnphcy

VisDA: The Visual Domain Adaptation Challenge [article]

Xingchao Peng, Ben Usman, Neela Kaushik, Judy Hoffman, Dequan Wang, Kate Saenko
2017 arXiv   pre-print
We present the 2017 Visual Domain Adaptation (VisDA) dataset and challenge, a large-scale testbed for unsupervised domain adaptation across visual domains.  ...  Our dataset is the largest one to date for cross-domain object classification, with over 280K images across 12 categories in the combined training, validation and testing domains.  ...  Majority of 3D model databases have thousands of rare classes with significantly unequal number of samples that are not present in the majority of other datasets that makes it difficult to use them  ... 
arXiv:1710.06924v2 fatcat:ylqhhypt3jecxpppvqlfcodeae

Source-Free Open Compound Domain Adaptation in Semantic Segmentation [article]

Yuyang Zhao, Zhun Zhong, Zhiming Luo, Gim Hee Lee, Nicu Sebe
2021 arXiv   pre-print
The model is evaluated on the samples from the target and unseen open domains.  ...  Furthermore, our model also achieves the leading performance on CityScapes for domain generalization.  ...  To reduce the large gap between synthetic data and the real-world data, DG methods usually augment the synthetic samples [38, 12] with the styles of ImageNet [7] or conditionally align the outputs  ... 
arXiv:2106.03422v1 fatcat:ffwvgxi4wjb5dcsgqa2skeer5i

Adaptive Semantic Segmentation with a Strategic Curriculum of Proxy Labels [article]

Kashyap Chitta, Jianwei Feng, Martial Hebert
2018 arXiv   pre-print
We conduct a series of experiments with the GTA5, Cityscapes and BDD100k datasets on synthetic-to-real domain adaptation and geographic domain adaptation, showing the advantages of our method over baselines  ...  Our architecture then allows selective mining of easy samples from this set of proxy labels, and hard samples from the annotated source domain.  ...  For improving unsupervised domain adaptation, we introduce three modifications: class-wise weighting of the loss function, easy mining in the proxy-labeled target domain samples, and hard mining in the  ... 
arXiv:1811.03542v1 fatcat:ieuw362knfhq7fj6xzhdtoggnu

Road images augmentation with synthetic traffic signs using neural networks [article]

Anton Konushin, Boris Faizov, Vlad Shakhuro
2021 arXiv   pre-print
We adapt a variational autoencoder for sampling plausible locations of new traffic signs in images.  ...  However, the state of the art methods works only for frequent sign classes, which are well represented in training datasets. We consider the task of rare traffic sign detection and classification.  ...  Proposed methods allow augmenting the real road images with high-quality synthetic traffic signs for classes, which are absent in the real training dataset.  ... 
arXiv:2101.04927v1 fatcat:v5qvswlskrfldeyhxnqzdjvpem

Road images augmentation with synthetic traffic signs using neural networks

A.S. Konushin, B.V. Faizov, V.I. Shakhuro
2021 Computer Optics  
We adapt a variational autoencoder for sampling plausible locations of new traffic signs in images.  ...  However, the state of the art methods works only for frequent sign classes, which are well represented in training datasets. We consider the task of rare traffic sign detection and classification.  ...  It showed that our method for generating synthetic training samples improved quality for detector and classifier of traffic signs. For all sign classes, recognition quality improved.  ... 
doi:10.18287/2412-6179-co-859 fatcat:grncol6wzvehlb5rcvgxfakery

Unravelling Small Sample Size Problems in the Deep Learning World [article]

Rohit Keshari, Soumyadeep Ghosh, Saheb Chhabra, Mayank Vatsa, Richa Singh
2020 arXiv   pre-print
For problems with large training databases, deep learning models have achieved superlative performances.  ...  However, there are a lot of small sample size or S^3 problems for which it is not feasible to collect large training databases.  ...  Data Augmentation For Small Sample Learning problems, the role of input space has been well explored in the literature of domain adaptation via data augmentation.  ... 
arXiv:2008.03522v1 fatcat:nigmkyma6rahvfhylcml3xmmxq

Fake It Till You Make It: Face analysis in the wild using synthetic data alone [article]

Erroll Wood, Tadas Baltrušaitis, Charlie Hewitt, Sebastian Dziadzio, Matthew Johnson, Virginia Estellers, Thomas J. Cashman, Jamie Shotton
2021 arXiv   pre-print
Researchers have tried to bridge this gap with data mixing, domain adaptation, and domain-adversarial training, but we show that it is possible to synthesize data with minimal domain gap, so that models  ...  The community has long enjoyed the benefits of synthesizing training data with graphics, but the domain gap between real and synthetic data has remained a problem, especially for human faces.  ...  Acknowledgements We thank Pedro Urbina, Jon Hanzelka, Rodney Brunet, and Panagiotis Giannakopoulos for their artistic contributions. This work was published in ICCV 2021.  ... 
arXiv:2109.15102v2 fatcat:m7z4lmddhzbdbbojkk7kbzilki
« Previous Showing results 1 — 15 out of 14,095 results