Approximating CNNs with Bag-of-local-Features models works surprisingly well on ImageNet
Wieland Brendel Matthias Bethge
Abstract
Deep Neural Networks (DNNs) excel on many complex perceptual tasks but it has proven notoriously difficult to understand how they reach their decisions. We here introduce a high-performance DNN architecture on ImageNet whose decisions are considerably easier to explain. Our model, a simple variant of the ResNet-50 architecture called BagNet, classifies an image based on the occurrences of small local image features without taking into account their spatial ordering. This strategy is closely related to the bag-of-feature (BoF) models popular before the onset of deep learning and reaches a surprisingly high accuracy on ImageNet (87.6% top-5 for 33 $\times$ 33 px features and Alexnet performance for 17 $\times$ 17 px features). The constraint on local features makes it straight-forward to analyse how exactly each part of the image influences the classification. Furthermore, the BagNets behave 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. This suggests that the improvements of DNNs over previous bag-of-feature classifiers in the last few years is mostl
中文速览
深度神经网络（DNN）在图像识别上表现出色，但它们"怎么做出决策"一直难以解释。研究者受深度学习兴起前流行的"词袋"模型（bag-of-features, BoF）启发，设计了一种名为 BagNet 的简化 ResNet-50 变体：它只依赖图像中小块局部区域（最小 17×17 像素）的特征出现频次来分类，完全不考虑这些特征的空间位置关系，因此每个图像区块对最终决策的贡献可以被精确追踪和可视化。实验表明，BagNet 在 ImageNet 上达到了令人意外的高精度（33×33 像素感受野时 top-5 准确率达 87.6%），同时其决策热图能清晰揭示模型依赖哪些局部纹理或形状线索，也能直观暴露误分类的原因。更关键的是，BagNet 与 VGG-16、ResNet-152、DenseNet-169 等主流高性能网络在特征敏感性、错误分布和图像块交互方式上高度相似，这暗示当前主流 DNN 的决策逻辑可能与 BoF 模型并无本质区别——近年来的精度提升更多来自更好的训练调优，而非学到了真正整体性的、能理解图像因果关系的特征。
原文 arXiv:1904.00760;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1904.00760v1