论文标题
形状文本辩护的神经网络培训
Shape-Texture Debiased Neural Network Training
论文作者
论文摘要
形状和纹理是识别对象的两个突出和互补提示。但是，根据训练数据集，卷积神经网络通常会偏向纹理或形状。我们的消融表明，这种偏见会退化模型性能。在这一观察过程中，我们开发了一种简单的算法，用于形状质量的偏见学习。为了防止模型在表示学习中的单个提示上专门参与，我们以形状和纹理信息相互冲突的图像（例如，柠檬纹的形状，但具有柠檬纹理的图像）来增强培训数据，最重要的是，同时提供了形状和纹理的相应监督。 实验表明，我们的方法成功地改善了几种图像识别基准和对抗性鲁棒性上的模型性能。例如，通过对Imagenet进行培训，它有助于Resnet-152对Imagenet（+1.2％），Imagenet-A（+5.2％），Imagenet-C（+8.3％）（+8.3％）和Stylized-Imagenet（+11.1％）以及针对FGSM对抗对FGSM的对抗攻击者的防御（+14.4.4％）的防御。我们的方法还声称与其他高级数据增强策略（例如，混合和cutmix）兼容。该代码可在此处提供：https：//github.com/liyingwei/shapetexturedebiasedTraining。
Shape and texture are two prominent and complementary cues for recognizing objects. Nonetheless, Convolutional Neural Networks are often biased towards either texture or shape, depending on the training dataset. Our ablation shows that such bias degenerates model performance. Motivated by this observation, we develop a simple algorithm for shape-texture debiased learning. To prevent models from exclusively attending on a single cue in representation learning, we augment training data with images with conflicting shape and texture information (eg, an image of chimpanzee shape but with lemon texture) and, most importantly, provide the corresponding supervisions from shape and texture simultaneously. Experiments show that our method successfully improves model performance on several image recognition benchmarks and adversarial robustness. For example, by training on ImageNet, it helps ResNet-152 achieve substantial improvements on ImageNet (+1.2%), ImageNet-A (+5.2%), ImageNet-C (+8.3%) and Stylized-ImageNet (+11.1%), and on defending against FGSM adversarial attacker on ImageNet (+14.4%). Our method also claims to be compatible with other advanced data augmentation strategies, eg, Mixup, and CutMix. The code is available here: https://github.com/LiYingwei/ShapeTextureDebiasedTraining.
