Corpus ID: 208175492

Exploring the Origins and Prevalence of Texture Bias in Convolutional Neural Networks

@article{Hermann2019ExploringTO,
  title={Exploring the Origins and Prevalence of Texture Bias in Convolutional Neural Networks},
  author={Katherine L. Hermann and Simon Kornblith},
  journal={ArXiv},
  year={2019},
  volume={abs/1911.09071}
}
Recent work has indicated that, unlike humans, ImageNet-trained CNNs tend to classify images by texture rather than shape. How pervasive is this bias, and where does it come from? We find that, when trained on datasets of images with conflicting shape and texture, the inductive bias of CNNs often favors shape; in general, models learn shape at least as easily as texture. Moreover, although ImageNet training leads to classifier weights that classify ambiguous images according to texture, shape… Expand
Shape or Texture: Understanding Discriminative Features in CNNs
TLDR
It is shown that a network learns the majority of overall shape information at the first few epochs of training and that this information is largely encoded in the last few layers of a CNN, as well as when the network learns about object shape during training. Expand
Assessing The Importance Of Colours For CNNs In Object Recognition
TLDR
It is demonstrated that CNNs often rely heavily on colour information while making a prediction, and the degree of dependency on colours tend to vary from one dataset to another. Expand
Permuted AdaIN: Enhancing the Representation of Local Cues in Image Classifiers
TLDR
This work provides a method that enhances the representation of local cues in the hidden layers of image classifiers, and consistently outperforms baseline methods in image classification, as well as in the setting of domain generalization. Expand
Shape Defense
TLDR
It is shown that a) edge information can also benefit other adversarial training methods, and b) CNNs trained on edge-augmented inputs are more robust against natural image corruptions such as motion blur, impulse noise and JPEG compression, than CNNstrained solely on RGB images. Expand
Shape Defense Against Adversarial Attacks
TLDR
This work explores how shape bias can be incorporated into CNNs to improve their robustness and shows that edge information can a) benefit other adversarial training methods, b) be even more effective in conjunction with background subtraction, c) be used to defend against poisoning attacks, and d) make CNNs more robust against natural image corruptions. Expand
Frustratingly Simple Domain Generalization via Image Stylization
TLDR
This work addresses the Domain Generalization problem, where the classifier must generalize to an unknown target domain, and demonstrates an extremely simple yet effective method, namely correcting this bias by augmenting the dataset with stylized images. Expand
Local features and global shape information in object classification by deep convolutional neural networks
TLDR
DCNNs' performance is an inversion of human performance with respect to global and local shape processing, Whereas abstract relations of elements predominate in human perception of shape, DCNNs appear to extract only local contour fragments, with no representation of how they spatially relate to each other to form global shapes. Expand
Reducing Domain Gap by Reducing Style Bias
TLDR
The Style-Agnostic Networks (SagNets) disentangle style encodings from class categories to prevent style biased predictions and focus more on the contents, which effectively reduces the style bias of CNNs and makes the model more robust under domain shift. Expand
Reducing Domain Gap via Style-Agnostic Networks
TLDR
Style-Agnostic Networks (SagNets) is proposed to encourage the model to focus more on image contents and ignore image styles, and win the 1st place in the semi-supervised domain adaptation task of the VisDA-2019 Challenge. Expand
Grounding deep neural network predictions of human categorization behavior in understandable functional features: The case of face identity
TLDR
It is shown that DNN predictions of human behavior improve because their shape and texture features overlap with those that subsume human behavior, which must equate the functional features that subsumes the behavioral performances of the brain and its models. Expand
...
1
2
...

References

SHOWING 1-10 OF 81 REFERENCES
ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness
TLDR
It is shown that ImageNet-trained CNNs are strongly biased towards recognising textures rather than shapes, which is in stark contrast to human behavioural evidence and reveals fundamentally different classification strategies. Expand
Deep convolutional networks do not classify based on global object shape
TLDR
Evidence is provided that DCNNs have access to some local shape information in the form of local edge relations, but they have no access to global object shapes. Expand
Do Better ImageNet Models Transfer Better?
TLDR
It is found that, when networks are used as fixed feature extractors or fine-tuned, there is a strong correlation between ImageNet accuracy and transfer accuracy, and ImageNet features are less general than previously suggested. Expand
Eigen-Distortions of Hierarchical Representations
TLDR
This work utilizes Fisher information to establish a model-derived prediction of sensitivity to local perturbations of an image, and finds that simple models of early visual processing, incorporating one or more stages of local gain control, trained on the same database of distortion ratings, provide substantially better predictions of human sensitivity than the CNN, or any combination of layers of VGG16. Expand
Generalisation in humans and deep neural networks
TLDR
The robustness of humans and current convolutional deep neural networks on object recognition under twelve different types of image degradations is compared and it is shown that DNNs trained directly on distorted images consistently surpass human performance on the exact distortion types they were trained on. Expand
Approximating CNNs with Bag-of-local-Features models works surprisingly well on ImageNet
TLDR
A high-performance DNN architecture on ImageNet whose decisions are considerably easier to explain is introduced, and behaves similar to state-of-the art deep neural networks such as VGG-16, ResNet-152 or DenseNet-169 in terms of feature sensitivity, error distribution and interactions between image parts. Expand
Unsupervised Representation Learning by Predicting Image Rotations
TLDR
This work proposes to learn image features by training ConvNets to recognize the 2d rotation that is applied to the image that it gets as input, and demonstrates both qualitatively and quantitatively that this apparently simple task actually provides a very powerful supervisory signal for semantic feature learning. Expand
Cognitive Psychology for Deep Neural Networks: A Shape Bias Case Study
TLDR
This work proposes to address the interpretability problem in modern DNNs using the rich history of problem descriptions, theories and experimental methods developed by cognitive psychologists to study the human mind, and demonstrates the capability of tools from cognitive psychology for exposing hidden computational properties of DNN's while concurrently providing us with a computational model for human word learning. Expand
Why do deep convolutional networks generalize so poorly to small image transformations?
TLDR
The results indicate that the problem of insuring invariance to small image transformations in neural networks while preserving high accuracy remains unsolved. Expand
Image Style Transfer Using Convolutional Neural Networks
TLDR
A Neural Algorithm of Artistic Style is introduced that can separate and recombine the image content and style of natural images and provide new insights into the deep image representations learned by Convolutional Neural Networks and demonstrate their potential for high level image synthesis and manipulation. Expand
...
1
2
3
4
5
...