Adversarial learning in CV基于深度学习的图像分类网络,大多是在精心制作的数据集下进行训练,并完成相应的部署,对于数据集之外的图像或稍加改造的图像,网络的识别能力往往会受到一定的影响。在此现象之下,对抗攻击(Adversarial Attack)开始加入到网络模型鲁棒性的考查之中。通过添加不同的噪声或对图像的某些区域进行一定的改造生成对抗样本,以此样本对网络模型进行攻击以达到混淆网络的目的,即对抗攻击。而添加的这些干扰信息,在人眼看来是没有任何区别的,但是对于网络模型而言,某些数值的变化便会引起“牵一发而动全身”的影响。这在实际应用中将是非常重大的判定失误,如果发生在安检、安防等领域,将会出现不可估量的问题。
Marin Dujmović, Gaurav Malhotra, Jeffrey S Bowers
eLife, (2020): 1-27
Deep convolutional neural networks (DCNNs) are frequently described as the best current models of human and primate vision. An obvious challenge to this claim is the existence of that fool DCNNs but are uninterpretable to humans. However, recent research has suggested that there...
Cited by0BibtexViews7DOI
0
0
NIPS 2020, (2020)
Our research advances vision-and-language representation learning by incorporating adversarial training in both pre-training and finetuning stages
Cited by0BibtexViews273
0
0
Samarth Sinha, Sayna Ebrahimi,Trevor Darrell
ICCV, pp.5971-5980, (2019)
We describe a pool-based semisupervised active learning algorithm that implicitly learns this sampling mechanism in an adversarial manner
Cited by84BibtexViews31DOI
0
0
Egor Zakharov,Aliaksandra Shysheya, Egor Burkov,Victor S. Lempitsky
ICCV, pp.9458-9467, (2019)
We present a system with such few-shot capability
Cited by72BibtexViews51DOI
0
0
CVPR, (2019): 1633-1642
We propose a novel method to address these problems
Cited by23BibtexViews43
0
0
international conference on learning representations, (2019)
We present SPIGAN, a novel method for leveraging synthetic data and Privileged Information available in simulated environments to perform unsupervised domain adaptation of deep networks
Cited by22BibtexViews40
0
0
IEEE transactions on medical imaging, no. 11 (2019): 1-1
To tackle the domain shift between the source and target domains, we exploited unsupervised domain adaptation model to improve the generalization of the segmentation network
Cited by21BibtexViews14DOI
0
0
CVPR, (2019): 3527-3536
We propose the decorrelated adversarial learning method for Age Invariant Face Recognition
Cited by9BibtexViews25
0
0
CVPR, pp.1476-1485, (2019)
We demonstrated that feedback adversarial learning – leveraging discriminator information into the feed-forward path of the generation process – is a simple yet effective method to improve existing generative adversarial frameworks
Cited by8BibtexViews74
0
0
CVPR, pp.11907-11916, (2019)
We identify three major factors: 1) appearance variations; 2) pose variations and 3) over-fitting issue with point estimation
Cited by7BibtexViews74
0
0
Pritish Uplavikar,Zhenyu Wu,Zhangyang Wang
CVPR Workshops, (2019): 1-8
Visualizing first two PCA components of the encoding Z learned by U-Net without adversarial loss. Colors points with the same water type, Colors points with the same content
Cited by5BibtexViews17
0
0
CVPR, pp.8415-8424, (2019)
We show that GCC-generative adversarial network yields better results compared to several state-of-the-art baselines for experiments involving human perception and image manipulation detection
Cited by2BibtexViews15
0
0
IEEE ACCESS, (2018): 14410.0-14430.0
The recent breakthrough in artificial intelligence in the form of tabula-rasa learning of AlphaGo Zero owes a fair share to deep Residual Networks that were originally proposed for the task of image recognition
Cited by473BibtexViews205DOI
1
0
computer and communications security, (2018): 2154-2156
We conclude this work by discussing some future research paths arising from the fact that machine learning has been originally developed for closed-world problems where the possible
Cited by345BibtexViews258DOI
0
0
Antonia Creswell, Tom White,Vincent Dumoulin, Kai Arulkumaran,Biswa Sengupta,Anil A. Bharath
IEEE Signal Processing Magazine, no. 1 (2018): 53-65
Diversity in the generator can be increased by practical hacks to balance the distribution of samples produced by the discriminator for real and fake batches, or by employing multiple G ENERATIVE adversarial networks to cover the different modes of the probability distribution
Cited by327BibtexViews51DOI
0
0
CVPR, (2018): 8990-8999
It enables the classifier to focus on the temporal robust features which are originally diminished during the training process
Cited by278BibtexViews54DOI
0
0
ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 31 (NIPS 2018), (2018): 3914-3924
We find that adversarial examples that transfer across computer vision models do successfully influence the perception of human observers, uncovering a new class of illusions that are shared between computer vision models and the human brain
Cited by212BibtexViews193
0
0
CVPR, (2018)
This paper has proposed an adversarial learning framework to transfer the 3D human pose structures learned from the fully annotated dataset to in-the-wild images with only 2D pose annotations
Cited by194BibtexViews81
0
0
CVPR, pp.9465-9474, (2018)
In this paper we proposed CartoonGAN, a Generative Adversarial Network to transform real-world photos to high-quality cartoon style images
Cited by124BibtexViews16
0
0
computer vision and pattern recognition, (2018)
We present a novel approach that combines Reinforcement Learning and Generative Adversarial Networks to generate more human-like responses to questions
Cited by77BibtexViews57DOI
0
0