Deep Residual Learning for Image Recognition--全论文翻译加解读

Abstract

摘要
           

      Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers—8× deeper than VGG nets [41] but still having lower complexity. An ensemble of these residual nets achieves 3.57% error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers.

      The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28% relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions1 , where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.

      深度神经网络很难训练。我们提出了一个残差学习框架，以简化比以前使用的网络深度大得多的网络的训练。我们明确地将层重新表述为参考层输入的学习残差函数，而不是学习未参考的函数。我们提供了全面的经验证据，表明这些残差网络更容易优化，并且可以从相当大的深度中获得精度。在ImageNet数据集上，我们评估了深度高达152层的剩余网络，比VGG网络[41]深8倍，但仍然具有较低的复杂性。这些残差网络的集合在ImageNet测试集上的误差达到3.57%。该结果在ILSVRC 2015分类任务中获得第一名。我们还介绍了100层和1000层的CIFAR-10分析。

      表征的深度对于许多视觉识别任务至关重要。仅由于我们的极深表示，我们在COCO对象检测数据集上获得了28%的相对改进。深度残差网络是我们参加ILSVRC & COCO 2015竞赛的基础1，我们还在ImageNet检测、ImageNet定位、COCO检测和COCO分割的任务中获得了第一名。

读者总结：

摘要部分展示了自己的成果，提出自己的工作是什么，并取得了哪些成果

1. Introduction

1、介绍

       Deep convolutional neural networks [22, 21] have led to a series of breakthroughs for image classification [21, 50, 40]. Deep networks naturally integrate low/mid/highlevel features [50] and classifiers in an end-to-end multilayer fashion, and the “levels” of features can be enriched by the number of stacked layers (depth). Recent evidence [41, 44] reveals that network depth is of crucial importance, and the leading results [41, 44, 13, 16] on the challenging ImageNet dataset [36] all exploit “very deep” [41] models, with a depth of sixteen [41] to thirty [16]. Many other nontrivial visual recognition tasks [8, 12, 7, 32, 27] have also greatly benefited from very deep models.

       深度卷积神经网络[22，21]为图像分类带来了一系列突破[21，50，40]。深度网络以端到端的多层方式自然地集成了低/中/高维度特征[50]和分类器，并且特征的“级别”可以通过堆叠层的数量（深度）来丰富。最近的证据[41，44]表明网络深度至关重要，在具有挑战性的ImageNet数据集[36]上的领先结果[41，44，13，16]都利用了“非常深”的[41]模型，深度为16[41]到30[16]。许多其他非平凡的视觉识别任务[8，12，7，32，27]也有极大地受益于非常深的模型。

      Driven by the significance of depth, a question arises: Is learning better networks as easy as stacking more layers? An obstacle to answering this question was the notorious problem of vanishing/exploding gradients [1, 9], which hamper convergence from the beginning. This problem, however, has been largely addressed by normalized initialization [23, 9, 37, 13] and intermediate normalization layers [16], which enable networks with tens of layers to start converging for stochastic gradient descent (SGD) with backpropagation [22].

         在深度的重要性的驱使下，一个问题出现了:学习更好的网络就像堆叠更多的层一样简单吗?回答这个问题的一个障碍是臭名昭著的梯度消失/爆炸问题[1,9]，它从一开始就阻碍了收敛。然而，这个问题已经通过规范化初始化[23,9,37,13]和中间规范化层[16]得到了很大程度的解决，这使得具有数十层的网络能够开始收敛随机梯度下降(SGD)与反向传播[22]。

          When deeper networks are able to start converging, a degradation problem has been exposed: with the network depth increasing, accuracy gets saturated (which might be unsurprising) and then degrades rapidly. Unexpectedly, such degradation is not caused by overfitting, and adding more layers to a suitably deep model leads to higher training error, as reported in [11, 42] and thoroughly verified by our experiments. Fig. 1 shows a typical example.

       当更深的网络能够开始收敛时，一个退化问题就暴露出来了:随着网络深度的增加，精度趋于饱和(这可能不足为奇)，然后迅速退化。出乎意料的是，这种退化并不是由过拟合引起的，在适当深度的模型上增加更多的层会导致更高的训练误差，这一点在[11,42]中有报道，我们的实验也充分验证了这一点。图1是一个典型的例子。

这里是图1：

 Figure 1. Training error (left) and test error (right) on CIFAR-10 with 20-layer and 56-layer “plain” networks. The deeper network has higher training error, and thus test error. Similar phenomena on ImageNet is presented in Fig. 4.

图1所示。CIFAR-10在20层和56层“普通”网络上的训练误差(左)和测试误差(右)。网络越深，训练误差越大，测试误差也越大。ImageNet上的类似现象如图4所示。

           The degradation (of training accuracy) indicates that not all systems are similarly easy to optimize. Let us consider a shallower architecture and its deeper counterpart that adds more layers onto it. There exists a solution by construction to the deeper model: the added layers are identity mapping, and the other layers are copied from the learned shallower model. The existence of this constructed solution indicates that a deeper model should produce no higher training error than its shallower counterpart. But experiments show that our current solvers on hand are unable to find solutions that are comparably good or better than the constructed solution (or unable to do so in feasible time).

        (训练精度的)退化表明，并非所有系统都同样容易优化。让我们考虑一个较浅的体系结构和它的更深的对应物，它在上面添加了更多的层。通过构造更深的模型存在一个解决方案:添加的层是身份映射，其他层是从学习的较浅模型复制的。这种构造解的存在表明，较深的模型不会比较浅的模型产生更高的训练误差。但实验表明，我们现有的解算器无法找到与构建的解决方案相当好或更好(或无法在可行的时间内做到这一点)。

           In this paper, we address the degradation problem by introducing a deep residual learning framework. Instead of hoping each few stacked layers directly fit a desired underlying mapping, we explicitly let these layers fit a residual mapping. Formally, denoting the desired underlying mapping as H(x), we let the stacked nonlinear layers fit another mapping of F(x) := H(x)−x. The original mapping is recast into F(x)+x. We hypothesize that it is easier to optimize the residual mapping than to optimize the original, unreferenced mapping. To the extreme, if an identity mapping were optimal, it would be easier to push the residual to zero than to fit an identity mapping by a stack of nonlinear layers

             在本文中，我们通过引入深度残差学习框架来解决退化问题。我们不是希望每几个堆叠层直接符合期望的底层映射，而是明确地让这些层符合残差映射。形式上，将期望的底层映射表示为H(x)，我们让堆叠的非线性层适合F(x)的另一个映射:= H(x) - x。原始映射被重铸为F(x)+x。我们假设优化残差映射比优化原始的、未引用的映射更容易。在极端情况下，如果一个恒等映射是最优的，将残差推至零要比用一堆非线性层拟合一个恒等映射容易得多

        The formulation of F(x) +x can be realized by feedforward neural networks with “shortcut connections” (Fig. 2).Shortcut connections [2, 34, 49] are those skipping one or more layers. In our case, the shortcut connections simply perform identity mapping, and their outputs are added to the outputs of the stacked layers (Fig. 2). Identity shortcut connections add neither extra parameter nor computational complexity. The entire network can still be trained end-to-end by SGD with backpropagation, and can be easily implemented using common libraries (e.g., Caffe [19]) without modifying the solvers.

      F(x) +x的表达式可以通过具有“快捷连接”的前馈神经网络来实现(图2)。快捷连接[2,34,49]是那些跳过一个或多个层的连接。在我们的例子中，快捷连接只是执行身份映射，它们的输出被添加到堆叠层的输出中(图2)。身份快捷连接既不增加额外的参数，也不增加计算复杂度。整个网络仍然可以通过反向传播的SGD进行端到端训练，并且可以使用通用库(例如Caffe[19])轻松实现，而无需修改求解器。

图2：

                                                 

         We present comprehensive experiments on ImageNet [36] to show the degradation problem and evaluate our method. We show that: 1) Our extremely deep residual nets are easy to optimize, but the counterpart “plain” nets (that simply stack layers) exhibit higher training error when the depth increases; 2) Our deep residual nets can easily enjoy accuracy gains from greatly increased depth, producing results substantially better than previous networks.

      我们在ImageNet上进行了全面的实验[36]，以显示退化问题并评估我们的方法。我们发现:1)我们的极深残差网络很容易优化，但对应的“普通”网络(简单地堆叠层)在深度增加时表现出更高的训练误差;2)我们的深度残差网络可以很容易地从深度大大增加中获得精度增益，产生的结果比以前的网络好得多。

      Similar phenomena are also shown on the CIFAR-10 set [20], suggesting that the optimization difficulties and the effects of our method are not just akin to a particular dataset.We present successfully trained models on this dataset with over 100 layers, and explore models with over 1000 layers.

        类似的现象也出现在CIFAR-10集上[20]，这表明我们的方法的优化困难和效果并不仅仅类似于特定的数据集。我们在超过100层的数据集上展示了成功训练的模型，并探索了超过1000层的模型。

          On the ImageNet classification dataset [36], we obtain excellent results by extremely deep residual nets. Our 152layer residual net is the deepest network ever presented on ImageNet, while still having lower complexity than VGG nets [41]. Our ensemble has 3.57% top-5 error on the ImageNet test set, and won the 1st place in the ILSVRC 2015 classification competition. The extremely deep representations also have excellent generalization performance on other recognition tasks, and lead us to further win the 1st places on: ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation in ILSVRC & COCO 2015 competitions. This strong evidence shows that the residual learning principle is generic, and we expect that it is applicable in other vision and non-vision problems. 

         在ImageNet分类数据集[36]上，我们使用极深残差网获得了很好的结果。我们的152层残差网络是迄今为止在ImageNet上呈现的最深的网络，但其复杂度仍低于VGG网络[41]。我们的集合在ImageNet测试集上的前5名错误率为3.57%，并在ILSVRC 2015分类大赛中获得第一名。极深表征在其他识别任务上也有出色的泛化性能，并使我们在ILSVRC & COCO 2015竞赛中，在ImageNet检测、ImageNet定位、COCO检测和COCO分割方面进一步获得第一名。这有力地证明了残差学习原理是通用的，我们期望它能适用于其他视觉和非视觉问题。

读者总结：

       图一中所展示的图为不参用残差网络的传统数据模型，在随着神经网络的深度由20层加深到56层，训练的误差反而要高于20层，这时的训练开销过大，导致最终结果偏离了最优解。真是这种情况的出现，所以作者引出了自己的工作，也就是图2的大致思路。

      假如我们的样本在10层左右便能够找到相对的最优解，那么此时其他的卷积层成为我们模型的负担，它们将耗费额外的时间，同时不一定能得到比先前更优的解。这时，就出现了退化现象，此时对于多余的网络层需要做到的就是令f(x)=x，忽略这些模型的训练结果，免去模型退化。这时，就出现了图2的残差模型。我们将输入x前馈到结果前，使得H（x）=f(x)+x，将x挪到左边，即为f(x)=H(x)-x，此时训练的就是输入与输入之间的差值，即残差，对残差进行微调，慢慢趋向最优解，如果不能的话，就将参数设为零，模型保持原有的解。

      这个模型解决的问题可以简述为“不一定使模型变得更好，但一定不能变得更坏”。对H（x）=f(x)+x求导后多加了一项。使得梯度更不容易因为连乘而消失。

2. Related Work

2、相关工作

         Residual Representations. In image recognition, VLAD[18] is a representation that encodes by the residual vectorswith respect to a dictionary, and Fisher Vector [30] can be
formulated as a probabilistic version [18] of VLAD. Bothof them are powerful shallow representations for image retrieval and classification [4, 48]. For vector quantization,
encoding residual vectors [17] is shown to be more effective than encoding original vectors

         残差陈述。在图像识别中，VLAD[18]是通过相对于字典的残差向量进行编码的表示，并且Fisher Vector[30]可以被公式化为VLAD的概率版本[18]。它们都是用于图像检索和分类的强大的浅层表示[4，48]。对于矢量量化，编码残差矢量[17]被证明比编码原始矢量更有效。

           In low-level vision and computer graphics, for solving Partial Differential Equations (PDEs), the widely used Multigrid method [3] reformulates the system as subproblems at multiple scales, where each subproblem is responsible for the residual solution between a coarser and a finer scale. An alternative to Multigrid is hierarchical basis preconditioning [45, 46], which relies on variables that represent residual vectors between two scales. It has been shown [3, 45, 46] that these solvers converge much faster than standard solvers that are unaware of the residual nature of the solutions. These methods suggest that a good reformulation or preconditioning can simplify the optimizatio

           在低级视觉和计算机图形学中，为了求解偏微分方程（PDE），广泛使用的多网格方法[3]将系统重新表述为多个尺度上的子问题，其中每个子问题负责粗尺度和细尺度之间的残差解。多重网格的一种替代方案是分层基础预处理[45，46]，它依赖于表示两个尺度之间的残差向量的变量。已经表明[3，45，46]，这些解算器比不知道解的残差性质的标准解算器收敛得更快。这些方法表明，良好的重新配方或预处理可以简化优化。

           Shortcut Connections. Practices and theories that lead to shortcut connections [2, 34, 49] have been studied for a long time. An early practice of training multi-layer perceptrons (MLPs) is to add a linear layer connected from the network input to the output [34, 49]. In [44, 24], a few intermediate layers are directly connected to auxiliary classifiers for addressing vanishing/exploding gradients. The papers of [39, 38, 31, 47] propose methods for centering layer responses, gradients, and propagated errors, implemented by shortcut connections. In [44], an “inception” layer is composed of a shortcut branch and a few deeper branches.

           快捷方式连接。导致捷径连接的实践和理论[2，34，49]已经研究了很长时间。训练多层感知器（MLP）的早期实践是添加从网络输入连接到输出的线性层[34，49]。在[44，24]中，一些中间层直接连接到辅助分类器，用于处理消失/爆炸梯度。[39，38，31，47]的论文提出了通过快捷连接实现的层响应、梯度和传播误差居中的方法。在[44]中，“初始”层由一个快捷分支和几个更深的分支组成。

                  Concurrent with our work, “highway networks” [42, 43] present shortcut connections with gating functions [15].These gates are data-dependent and have parameters, in contrast to our identity shortcuts that are parameter-free.When a gated shortcut is “closed” (approaching zero), the layers in highway networks represent non-residual functions. On the contrary, our formulation always learns residual functions; our identity shortcuts are never closed, and all information is always passed through, with additional residual functions to be learned. In addition, high-way networks have not demonstrated accuracy gains with extremely increased depth (e.g., over 100 layers).

            与我们的工作同时，“高速公路网络”[42，43]提供了具有门控功能的快捷连接[15]。

这些门是数据相关的，并且有参数，与我们的无参数身份快捷方式形成对比。

当门控快捷方式“关闭”（接近零）时，高速公路网络中的层表示非残差函数。相反，我们的公式总是学习残差函数；我们的身份快捷方式永远不会关闭，所有信息总是通过，还有额外的剩余功能需要学习。此外-高速公路方式网络还没有表现出随着深度的极大增加（例如超过100层）而获得的精度增益。

读者总结

相关工作板块点名了残差学习在处理梯度消失爆炸，不会丢失原有的精度，以及稳定提高的优点等。

3. Deep Residual Learning

3. Deep Residual Learning

3.1. Residual Learning

3.1 残差学习

              Let us consider H(x) as an underlying mapping to be fit by a few stacked layers (not necessarily the entire net), with x denoting the inputs to the first of these layers. If one hypothesizes that multiple nonlinear layers can asymptotically approximate complicated functions , then it is equivalent to hypothesize that they can asymptotically approximate the residual functions, i.e., H(x) − x (assuming that the input and output are of the same dimensions). So rather than expect stacked layers to approximate H(x), we explicitly let these layers approximate a residual function F(x) := H(x) − x. The original function thus becomes F(x)+x. Although both forms should be able to asymptotically approximate the desired functions (as hypothesized), the ease of learning might be different.

让我们将H（x）视为由几个堆叠层（不一定是整个网络）拟合的底层映射，其中x表示这些层中第一层的输入。如果假设多个非线性层可以渐近逼近复杂函数，那么等价于假设它们可以渐近逼近残差函数，即H（x）−x（假设输入和输出具有相同的维度）。因此，我们不期望堆叠层近似于H（x），而是明确地让这些层近似于残差函数F（x）：=H（x）−x。因此，原始函数变为F（x）+x。尽管这两种形式都应该能够渐近逼近所需的函数（如假设的那样），但学习的容易程度可能不同。

This reformulation is motivated by the counterintuitive phenomena about the degradation problem (Fig. 1, left). As we discussed in the introduction, if the added layers can be constructed as identity mappings, a deeper model should have training error no greater than its shallower counterpart. The degradation problem suggests that the solvers might have difficulties in approximating identity mappings by multiple nonlinear layers. With the residual learning reformulation, if identity mappings are optimal, the solvers may simply drive the weights of the multiple nonlinear layers toward zero to approach identity mappings.

这种重新表述的动机是关于退化问题的违反直觉的现象（图1，左）。正如我们在引言中所讨论的，如果添加的层可以被构造为身份映射，那么更深的模型的训练误差应该不大于其较浅的对应模型。退化问题表明，求解器可能难以通过多个非线性层来近似单位映射。通过残差学习重新表述，如果单位映射是最优的，则解算器可以简单地将多个非线性层的权重推向零，以接近单位映射。

In real cases, it is unlikely that identity mappings are optimal, but our reformulation may help to precondition the problem. If the optimal function is closer to an identity mapping than to a zero mapping, it should be easier for the solver to find the perturbations with reference to an identity mapping, than to learn the function as a new one. We show by experiments (Fig. 7) that the learned residual functions in general have small responses, suggesting that identity mappings provide reasonable preconditioning.

在实际情况下，身份映射不太可能是最优的，但我们的重新表述可能有助于解决问题。如果最优函数更接近恒等映射而不是零映射，那么求解器应该更容易找到关于恒等映射的扰动，而不是将函数学习为新函数。我们通过实验表明（图7），学习的残差函数通常具有较小的响应，这表明身份映射提供了合理的预处理。

3.2. Identity Mapping by Shortcuts

3.2.通过快捷方式进行身份映射

We adopt residual learning to every few stacked layers.

A building block is shown in Fig. 2. Formally, in this paper we consider a building block defined as:

我们对每堆叠几层采用残差学习。构建块如图2所示。形式上，在本文中，我们考虑的构建块定义为：

             Here x and y are the input and output vectors of the layers considered. The function F(x; fWig) represents the residual mapping to be learned. For the example in Fig. 2 that has two layers, F = W2σ(W1x) in which σ denotes ReLU [29] and the biases are omitted for simplifying notations. The operation F + x is performed by a shortcut connection and element-wise addition. We adopt the second nonlinearity after the addition (i.e., σ(y), see Fig. 2).

这里，x和y是所考虑的层的输入和输出向量。函数F（x；fWig）表示要学习的残差映射。对于图2中有两层的例子，F=W2σ（W1x），其中σ表示ReLU[29]和偏差被省略以简化符号。操作F+x是通过快捷连接和元素相加来执行的。我们采用加法后的第二个非线性（即σ（y），见图2）。

              The shortcut connections in Eqn.(1) introduce neither extra parameter nor computation complexity. This is not only attractive in practice but also important in our comparisons between plain and residual networks. We can fairly compare plain/residual networks that simultaneously have the same number of parameters, depth, width, and computational cost (except for the negligible element-wise addition).

方程中的快捷连接。（1） 既不引入额外的参数也不引入计算复杂性。这不仅在实践中很有吸引力，而且在我们比较纯网络和残差网络时也很重要。我们可以公平地比较同时具有相同数量的参数、深度、宽度和计算成本的纯/残差网络（除了可忽略的元素相加）。

           The dimensions of x and F must be equal in Eqn.(1).If this is not the case (e.g., when changing the input/output channels), we can perform a linear projection Ws by the shortcut connections to match the dimensions:

在等式中，x和F的尺寸必须相等。（1） 。如果不是这种情况（例如，当改变输入/输出通道时），我们可以通过快捷连接执行线性投影Ws，以匹配尺寸：

 3.3. Network Architectures

3.3网络架构

We have tested various plain/residual nets, and have observed consistent phenomena. To provide instances for discussion, we describe two models for ImageNet as follows.

Plain Network. Our plain baselines (Fig. 3, middle) are mainly inspired by the philosophy of VGG nets [41] (Fig. 3, left). The convolutional layers mostly have 3×3 filters and follow two simple design rules: (i) for the same output feature map size, the layers have the same number of filters; and (ii) if the feature map size is halved, the number of filters is doubled so as to preserve the time complexity per layer. We perform downsampling directly by convolutional layers that have a stride of 2. The network ends with a global average pooling layer and a 1000-way fully-connected layer with softmax. The total number of weighted layers is 34 in Fig. 3 (middle).

我们测试了各种平原/残余网，并观察到了一致的现象。为了提供讨论的实例，我们描述了ImageNet的两个模型，如下所示。

普通网络。我们的简单基线（图3，中间）主要受到VGG网络[41]哲学的启发（图3（左））。卷积层大多具有3×3个滤波器，并遵循两个简单的设计规则：（i）对于相同的输出特征图大小，层具有相同数量的滤波器；以及（ii）如果特征图大小减半，则滤波器的数量加倍，以保持每层的时间复杂度。我们通过步长为2的卷积层直接执行下采样。网络以全局平均池化层和具有softmax的1000路全连接层结束。在图3（中间）中，加权层的总数为34。

It is worth noticing that our model has fewer filters and lower complexity than VGG nets [41] (Fig. 3, left). Our 34layer baseline has 3.6 billion FLOPs (multiply-adds), which is only 18% of VGG-19 (19.6 billion FLOPs).

值得注意的是，我们的模型比VGG网络[41]具有更少的滤波器和更低的复杂性（图3，左）。我们的34层基线有36亿FLOP（乘以-

图3：ImageNet的网络架构示例。左图：VGG-19模型[41]（196亿FLOP）作为参考。中间：具有34个参数层（36亿FLOP）的平面网络。

右图：具有34个参数层（36亿FLOP）的残差网络。虚线快捷方式增加了尺寸。表1显示了更多细节和其他变体。

Residual Network. Based on the above plain network, we insert shortcut connections (Fig. 3, right) which turn the network into its counterpart residual version. The identity shortcuts (Eqn.(1)) can be directly used when the input and output are of the same dimensions (solid line shortcuts in Fig. 3). When the dimensions increase (dotted line shortcuts in Fig. 3), we consider two options: (A) The shortcut still performs identity mapping, with extra zero entries padded for increasing dimensions. This option introduces no extra parameter; (B) The projection shortcut in Eqn.(2) is used to match dimensions (done by 1×1 convolutions). For both options, when the shortcuts go across feature maps of two sizes, they are performed with a stride of 2.

剩余网络。基于上述普通网络，我们插入快捷连接（图3，右），将网络转换为其对应的残差版本。当输入和输出具有相同尺寸时，可以直接使用身份快捷方式（等式（1））（图3中的实线快捷方式）。当维度增加时（图3中的虚线快捷方式），我们考虑两个选项：（A）快捷方式仍然执行身份映射，为增加维度填充额外的零项。此选项不引入额外的参数；（B） 方程中的投影快捷方式。（2） 用于匹配维度（通过1×1卷积完成）。对于这两个选项，当快捷方式跨越两种尺寸的要素图时，它们将以2的步幅执行。

 3.4. Implementation

3.4 实施

Our implementation for ImageNet follows the practice in [21, 41]. The image is resized with its shorter side randomly sampled in [256; 480] for scale augmentation [41].

A 224×224 crop is randomly sampled from an image or its horizontal flip, with the per-pixel mean subtracted [21]. The standard color augmentation in [21] is used. We adopt batch normalization (BN) [16] right after each convolution and before activation, following [16]. We initialize the weights as in [13] and train all plain/residual nets from scratch. We use SGD with a mini-batch size of 256. The learning rate starts from 0.1 and is divided by 10 when the error plateaus, and the models are trained for up to 60 × 104 iterations. We use a weight decay of 0.0001 and a momentum of 0.9. We do not use dropout [14], following the practice in [16].

In testing, for comparison studies we adopt the standard 10-crop testing [21]. For best results, we adopt the fullyconvolutional form as in [41, 13], and average the scores at multiple scales (images are resized such that the shorter side is in f224; 256; 384; 480; 640g).

我们对ImageNet的实现遵循了[21，41]中的实践。调整图像的大小，使其短边在[256；480]中随机采样，以进行比例放大[41]。

从图像或其水平翻转中随机采样224×224裁剪，减去每个像素的平均值[21]。使用[21]中的标准增色。我们在每次卷积之后和激活之前采用批量归一化（BN）[16]，如下[16]。我们如[13]中所述初始化权重，并从头开始训练所有平网/残差网。我们使用小批量大小为256的SGD。学习率从0.1开始，当误差稳定时除以10，并且模型被训练多达60×104次迭代。我们使用0.0001的重量衰减和0.9的动量。按照[16]中的做法，我们不使用dropout[14]。

在测试中，为了进行比较研究，我们采用了标准的10作物测试[21]。为了获得最佳结果，我们采用了[41，13]中的完全卷积形式，并在多个尺度上对分数进行平均（调整图像大小，使短边处于f224；256；384；480；640g）。

读者总结：

（i）对于相同的输出特征图尺寸，层具有相同数量的滤波器；
（ii）如果特征图尺寸减半，则滤波器数量加倍，以便保持每层的时间复杂度。我们通过步长为2的卷积层直接执行下采样。

（iii）对于恒等映射层，为了保持特征图大小一致，我们可以用0来填充，使大小一致。

（iiii）为了保持通道数一致，我们可以通过1×1的卷积核来使得通道数一致

4. Experiments

4、实验

4.1. ImageNet Classification

4.1 ImageNet分类

We evaluate our method on the ImageNet 2012 classification dataset [36] that consists of 1000 classes. The models are trained on the 1.28 million training images, and evaluated on the 50k validation images. We also obtain a final result on the 100k test images, reported by the test server.We evaluate both top-1 and top-5 error rates.

我们在ImageNet 2012分类数据集[36]上评估了我们的方法，该数据集由1000个类组成。模型在128万张训练图像上进行训练，并在5万张验证图像上进行评估。我们还获得了测试服务器报告的10万张测试图像的最终结果。

我们评估了前1名和前5名的错误率。

Plain Networks. We first evaluate 18-layer and 34-layer plain nets. The 34-layer plain net is in Fig. 3 (middle). The 18-layer plain net is of a similar form. See Table 1 for detailed architectures.

The results in Table 2 show that the deeper 34-layer plain net has higher validation error than the shallower 18-layer plain net. To reveal the reasons, in Fig. 4 (left) we compare their training/validation errors during the training procedure. We have observed the degradation problem - the

34-layer plain net has higher training error throughout the whole training procedure, even though the solution space of the 18-layer plain network is a subspace of that of the 34-layer one.

We argue that this optimization difficulty is unlikely to be caused by vanishing gradients. These plain networks are trained with BN [16], which ensures forward propagated signals to have non-zero variances. We also verify that the backward propagated gradients exhibit healthy norms with BN. So neither forward nor backward signals vanish. In fact, the 34-layer plain net is still able to achieve competitive accuracy (Table 3), suggesting that the solver works to some extent. We conjecture that the deep plain nets may have exponentially low convergence rates, which impact the reducing of the training error3 . The reason for such optimization difficulties will be studied in the future.

普通网络。我们首先评估了18层和34层平面网。34层素网如图3（中）所示。18层的平面网也是类似的形式，具体架构见表1。

表2中的结果表明，较深的34层平面网比较浅的18层平面网具有更高的验证误差。为了揭示原因，在图4（左）中，我们比较了他们在训练过程中的训练/验证错误。我们观察到了退化问题——

尽管18层平面网络的解空间是34层平面网络解空间的子空间，但34层平面网在整个训练过程中具有更高的训练误差。

我们认为，这种优化困难不太可能是由梯度消失引起的。这些普通网络是用BN[16]训练的，这确保了前向传播的信号具有非零方差。我们还验证了后向传播的梯度与BN表现出健康的范数。因此，前向和后向信号都不会消失。事实上，34层平面网仍然能够达到有竞争力的精度（表3），这表明求解器在一定程度上起作用。我们推测，深平面网可能具有指数级低的收敛率，这影响了训练误差的降低3。这种优化困难的原因将在未来进行研究。

图像给出了每层的构造方法与和原神经网络的对比。

Residual Networks. Next we evaluate 18-layer and 34layer residual nets (ResNets). The baseline architectures are the same as the above plain nets, expect that a shortcut connection is added to each pair of 3×3 filters as in Fig. 3 (right). In the first comparison (Table 2 and Fig. 4 right), we use identity mapping for all shortcuts and zero-padding for increasing dimensions (option A). So they have no extra parameter compared to the plain counterparts.

We have three major observations from Table 2 and Fig. 4. First, the situation is reversed with residual learning – the 34-layer ResNet is better than the 18-layer ResNet (by 2.8%). More importantly, the 34-layer ResNet exhibits considerably lower training error and is generalizable to the validation data. This indicates that the degradation problem is well addressed in this setting and we manage to obtain accuracy gains from increased depth

剩余网络。接下来我们评估18层和34层残差网（ResNets）。基线架构与上述普通网络相同，只是向每对3×3滤波器添加了一个快捷连接，如图所示。第3（右）段。在第一次比较中（表2和图4右），我们对所有快捷方式使用同一映射，对增加维度使用零填充（选项A）。因此，与普通的同类产品相比，它们没有额外的参数。

我们从表2和图4中得到了三个主要观察结果。首先，残差学习扭转了这种情况——34层ResNet比18层ResNet好（2.8%）。更重要的是，34层ResNet表现出相当低的训练误差，并且可以推广到验证数据中。这表明退化问题在这种情况下得到了很好的解决，我们设法从增加的深度中获得了精度增益

Second, compared to its plain counterpart, the 34-layerResNet reduces the top-1 error by 3.5% (Table 2), resulting from the successfully reduced training error (Fig. 4 right vs.left). This comparison verifies the effectiveness of residual learning on extremely deep systems.

第二，与平原相对应，34层 ResNet将前1名的误差减少了3.5%（表2），这是由于成功地减少了训练误差（图4右vs。左）。这种比较验证了残差学习在极深系统上的有效性。

Last, we also note that the 18-layer plain/residual nets are comparably accurate (Table 2), but the 18-layer ResNet converges faster (Fig. 4 right vs. left). When the net is “not overly deep” (18 layers here), the current SGD solver is still able to find good solutions to the plain net. In this case, the ResNet eases the optimization by providing faster convergence at the early stage.

最后，我们还注意到，18层纯网/残差网相对准确（表2），但18层ResNet收敛更快（图4右侧与左侧）。当网络“不太深”（此处为18层）时，当前的SGD求解器仍然能够找到普通网络的良好解决方案。在这种情况下，ResNet通过在早期阶段提供更快的收敛来简化优化。

Identity vs. Projection Shortcuts. We have shown that parameter-free, identity shortcuts help with training. Next we investigate projection shortcuts (Eqn.(2)). In Table 3 we compare three options: (A) zero-padding shortcuts are used for increasing dimensions, and all shortcuts are parameterfree (the same as Table 2 and Fig. 4 right); (B) projection shortcuts are used for increasing dimensions, and other shortcuts are identity; and (C) all shortcuts are projections.

Table 3 shows that all three options are considerably better than the plain counterpart. B is slightly better than A. We argue that this is because the zero-padded dimensions in A indeed have no residual learning. C is marginally better than B, and we attribute this to the extra parameters introduced by many (thirteen) projection shortcuts. But the small differences among A/B/C indicate that projection shortcuts are not essential for addressing the degradation problem. So we do not use option C in the rest of this paper, to reduce memory/time complexity and model sizes. Identity shortcuts are particularly important for not increasing the complexity of the bottleneck architectures that are introduced below

标识与投影快捷方式。我们已经证明 无参数、身份快捷方式有助于训练。接下来我们研究投影快捷方式（方程。（2））。在表3中，我们比较了三种选项：（A）零填充快捷方式用于增加维度，并且所有快捷方式都是无参数的（与表2和图4相同）；（B） 投影快捷方式用于增加维度，其他快捷方式为同一快捷方式；以及（C）所有快捷方式都是投影。

表3显示，这三种方案都比普通方案要好得多。B略好于A。我们认为这是因为A中的零填充维度确实没有剩余学习。C略好于B，我们将其归因于许多（十三）投影快捷方式引入的额外参数。但A/B/C之间的微小差异表明，投影快捷方式对于解决退化问题并不重要。因此，我们在本文的其余部分中不使用选项C，以减少内存/时间复杂性和模型大小。

Deeper Bottleneck Architectures. Next we describe our deeper nets for ImageNet. Because of concerns on the training time that we can afford, we modify the building block as a bottleneck design4 . For each residual function F, we use a stack of 3 layers instead of 2 (Fig. 5). The three layers are 1×1, 3×3, and 1×1 convolutions, where the 1×1 layers are responsible for reducing and then increasing (restoring) dimensions, leaving the 3×3 layer a bottleneck with smaller input/output dimensions. Fig. 5 shows an example, where both designs have similar time complexity.

The parameter-free identity shortcuts are particularly important for the bottleneck architectures. If the identity shortcut in Fig. 5 (right) is replaced with projection, one can show that the time complexity and model size are doubled, as the shortcut is connected to the two high-dimensional ends. So identity shortcuts lead to more efficient models for the bottleneck designs.

50-layer ResNet: We replace each 2-layer block in the34-layer net with this 3-layer bottleneck block, resulting in a 50-layer ResNet (Table 1). We use option B for increasing dimensions. This model has 3.8 billion FLOPs.

更深层的瓶颈架构。接下来，我们将为ImageNet描述我们更深层次的网络。由于担心我们能负担得起的训练时间，我们修改了构建块作为瓶颈设计4。对于每个残差函数F，我们使用3层的堆栈，而不是2层（图5）。这三层是1×1、3×3和1×1卷积，其中1×1层负责减少然后增加（恢复）维度，使3×3层成为输入/输出维度较小的瓶颈。图5显示了一个示例，其中两种设计具有相似的时间复杂性。

无参数身份快捷方式对于瓶颈体系结构来说尤其重要。如果将图5（右）中的身份快捷方式替换为投影，则可以显示时间复杂性和模型大小加倍，因为快捷方式连接到两个高维端。因此，身份快捷方式为瓶颈设计带来了更有效的模型。

50层ResNet：我们在中替换每个2层块34层网络与此3层瓶颈块，产生50层ResNet（表1）。我们使用选项B来增加尺寸。该模型有38亿FLOP。

101-layer and 152-layer ResNets: We construct 101layer and 152-layer ResNets by using more 3-layer blocks (Table 1). Remarkably, although the depth is significantly increased, the 152-layer ResNet (11.3 billion FLOPs) still has lower complexity than VGG-16/19 nets (15.3/19.6 billion FLOPs).

The 50/101/152-layer ResNets are more accurate than the 34-layer ones by considerable margins (Table 3 and 4).

We do not observe the degradation problem and thus enjoy significant accuracy gains from considerably increased depth. The benefits of depth are witnessed for all evaluation metrics (Table 3 and 4).

Comparisons with State-of-the-art Methods. In Table 4 we compare with the previous best single-model results.

Our baseline 34-layer ResNets have achieved very competitive accuracy. Our 152-layer ResNet has a single-model top-5 validation error of 4.49%. This single-model result outperforms all previous ensemble results (Table 5). We combine six models of different depth to form an ensemble (only with two 152-layer ones at the time of submitting).

This leads to 3.57% top-5 error on the test set (Table 5).

This entry won the 1st place in ILSVRC 2015.

101层和152层ResNets：我们通过使用更多的3层块来构建101层和152-ResNets（表1）。值得注意的是，尽管深度显著增加，但152层ResNet（113亿FLOP）的复杂性仍然低于VGG-16/19网络（153/196亿FLOPs）。

50/101/152层的ResNets比34层的ResNets更准确，差距很大（表3和4）。

我们没有观察到退化问题，因此从显著增加的深度中获得了显著的精度增益。所有评估指标都见证了深度的好处（表3和表4）。

与最先进方法的比较。在表4中，我们与之前的最佳单模型结果进行了比较。

我们的基线34层ResNets已经实现了极具竞争力的准确性。我们的152层ResNet的单个模型前5名验证误差为4.49%。该单个模型结果优于之前的所有集成结果（表5）。我们将六个不同深度的模型组合成一个集合（在提交时只有两个152层的模型）。

这导致测试集前五名误差为3.57%（表5）。

该参赛作品在2015年ILSVRC中获得第一名。

4.2. CIFAR-10 and Analysis

CIFAR-10及其分析

We conducted more studies on the CIFAR-10 dataset [20], which consists of 50k training images and 10k testing images in 10 classes. We present experiments trained on the training set and evaluated on the test set. Our focus is on the behaviors of extremely deep networks, but not on pushing the state-of-the-art results, so we intentionally use simple architectures as follows.

The plain/residual architectures follow the form in Fig. 3 (middle/right). The network inputs are 32×32 images, with the per-pixel mean subtracted. The first layer is 3×3 convolutions. Then we use a stack of 6n layers with 3×3 convolutions on the feature maps of sizes f32; 16; 8g respectively, with 2n layers for each feature map size. The numbers of filters are f16; 32; 64g respectively. The subsampling is performed by convolutions with a stride of 2. The network ends with a global average pooling, a 10-way fully-connected layer, and softmax. There are totally 6n+2 stacked weighted layers. The following table summarizes the architecture:

我们对CIFAR-10数据集[20]进行了更多的研究，该数据集由10个类别的50k个训练图像和10k个测试图像组成。我们展示了在训练集上训练的实验，并在测试集上进行了评估。我们关注的是极深度网络的行为，而不是推动最先进的结果，因此我们有意使用如下简单的架构。

平面/残差架构遵循图3（中/右）中的形式。网络输入是32×32个图像，减去每个像素的平均值。第一层是3×3卷积。然后，我们在大小为f32的特征图上使用具有3×3卷积的6n层的堆栈；16；每个特征图大小具有2n层。滤波器的数量为f16；32；64克。通过步长为2的卷积来执行子采样。网络以全局平均池、10路全连接层和softmax结束。总共有6n+2个堆叠的加权层。下表汇总

When shortcut connections are used, they are connected to the pairs of 3×3 layers (totally 3n shortcuts). On this dataset we use identity shortcuts in all cases (i.e., option A),

so our residual models have exactly the same depth, width, and number of parameters as the plain counterparts.

We use a weight decay of 0.0001 and momentum of 0.9, and adopt the weight initialization in [13] and BN [16] but with no dropout. These models are trained with a minibatch size of 128 on two GPUs. We start with a learning rate of 0.1, divide it by 10 at 32k and 48k iterations, and terminate training at 64k iterations, which is determined on a 45k/5k train/val split. We follow the simple data augmentation in [24] for training: 4 pixels are padded on each side, and a 32×32 crop is randomly sampled from the padded image or its horizontal flip. For testing, we only evaluate the single view of the original 32×32 image.

We compare n = f3; 5; 7; 9g, leading to 20, 32, 44, and 56-layer networks. Fig. 6 (left) shows the behaviors of the plain nets. The deep plain nets suffer from increased depth, and exhibit higher training error when going deeper. This phenomenon is similar to that on ImageNet (Fig. 4, left) and on MNIST (see [42]), suggesting that such an optimization difficulty is a fundamental problem.

Fig. 6 (middle) shows the behaviors of ResNets. Also similar to the ImageNet cases (Fig. 4, right), our ResNets manage to overcome the optimization difficulty and demonstrate accuracy gains when the depth increases.

We further explore n = 18 that leads to a 110-layer ResNet. In this case, we find that the initial learning rate of 0.1 is slightly too large to start converging5 . So we use 0.01 to warm up the training until the training error is below 80% (about 400 iterations), and then go back to 0.1 and continue training. The rest of the learning schedule is as done previously. This 110-layer network converges well (Fig. 6, middle). It has fewer parameters than other deep and thin networks such as FitNet [35] and Highway [42] (Table 6), yet is among the state-of-the-art results (6.43%, Table 6).

当使用快捷方式连接时，它们连接到3×3层的对（总共3n个快捷方式）。在这个数据集上，我们在所有情况下都使用身份快捷方式（即选项A）， 

因此，我们的残差模型具有与普通模型完全相同的深度、宽度和参数数量。

我们使用0.0001的权重衰减和0.9的动量，并采用[13]和BN[16]中的权重初始化，但没有遗漏。这些模型在两个GPU上使用128个小批量进行训练。我们从0.1的学习率开始，在32k和48k迭代时将其除以10，并在64k迭代时终止训练，这是在45k/5k训练/val分割上确定的。我们遵循[24]中的简单数据扩充进行训练：每侧填充4个像素，并从填充图像或其水平翻转中随机采样32×32裁剪。为了测试，我们只评估原始32×32图像的单个视图。

我们比较n＝f3；5.7.9g，导致20、32、44和56层网络。图6（左）显示了普通网的行为。深平原网的深度增加，并且当深入时表现出更高的训练误差。这种现象与ImageNet上的现象类似（图4，左）和MNIST上的现象（见[42]），表明这种优化困难是一个根本问题。

图6（中）显示了ResNets的行为。同样类似于ImageNet的情况（图4，右），当深度增加时，我们的ResNets设法克服了优化困难，并证明了准确性的提高。

我们进一步探索了n＝18，这导致了110层的ResNet。在这种情况下，我们发现0.1的初始学习率稍大，无法开始转换5。因此，我们使用0.01来预热训练，直到训练误差低于80%（约400次迭代），然后返回0.1并继续训练。学习计划的其余部分与之前一样。这个110层的网络该110层网络收敛良好（图6，中间）。它的参数比其他的深和细网络，如FitNet[35]和Highway[42]（表6），但属于最先进的结果之一（6.43%，表6）

Analysis of Layer Responses. Fig. 7 shows the standard deviations (std) of the layer responses. The responses are the outputs of each 3×3 layer, after BN and before other nonlinearity (ReLU/addition). For ResNets, this analysis reveals the response strength of the residual functions.

Fig. 7 shows that ResNets have generally smaller responses than their plain counterparts. These results support our basic motivation (Sec.3.1) that the residual functions might be generally closer to zero than the non-residual functions.

We also notice that the deeper ResNet has smaller magnitudes of responses, as evidenced by the comparisons among ResNet-20, 56, and 110 in Fig. 7. When there are more layers, an individual layer of ResNets tends to modify the signal less.

Exploring Over 1000 layers. We explore an aggressively deep model of over 1000 layers. We set n = 200 that leads to a 1202-layer network, which is trained as described above. Our method shows no optimization difficulty, and this 103 -layer network is able to achieve training error <0.1% (Fig. 6, right). Its test error is still fairly good (7.93%, Table 6).

But there are still open problems on such aggressively deep models. The testing result of this 1202-layer network is worse than that of our 110-layer network, although both have similar training error. We argue that this is because of overfitting. The 1202-layer network may be unnecessarily large (19.4M) for this small dataset. Strong regularization such as maxout [10] or dropout [14] is applied to obtain the best results ([10, 25, 24, 35]) on this dataset. In this paper, we use no maxout/dropout and just simply impose regularization via deep and thin architectures by design, without distracting from the focus on the difficulties of optimization. But combining with stronger regularization may improve results, which we will study in the future.

层响应分析。图7显示了层响应的标准偏差（std）。响应是BN之后和其他非线性（ReLU/加法）之前每个3×3层的输出。对于ResNets，该分析揭示了残差函数的响应强度。

图7显示，ResNets通常比普通的对应物具有更小的响应。这些结果支持了我们的基本动机（第3.1节），即残差函数通常可能比非残差函数更接近于零。

我们还注意到，更深的ResNet具有较小的响应幅度，如图中ResNet-20、56和110之间的比较所证明的。7。当有更多的层时，ResNets的单个层倾向于较少地修改信号。

探索超过1000层。我们探索了一个超过1000层的深度模型。我们设置n＝200，这导致如上所述训练的1202层网络。我们的方法没有显示出优化的困难，并且这个103层网络能够实现训练误差＜0.1%（图6，右）。其测试误差仍然相当好（7.93%，表6）。

但在如此激进的深度模型上仍然存在悬而未决的问题。这个1202层网络的测试结果比我们的110层网络更差，尽管两者有类似的训练错误。我们认为这是因为过度拟合。对于这个小数据集来说，1202层网络可能是不必要的大（19.4M）。应用诸如maxout[10]或dropout[14]之类的强正则化来在该数据集上获得最佳结果（[10，25，24，35]）。在本文中，我们不使用最大值/丢弃，只是通过设计通过深度和精简架构简单地强制正则化，而不会分散对优化困难的关注。但与更强的正则化相结合可能会改善结果，我们将在未来进行研究。

4.3. Object Detection on PASCAL and MS COCO

4.3.PASCAL和MS COCO上的目标检测

Our method has good generalization performance on other recognition tasks. Table 7 and 8 show the object detection baseline results on PASCAL VOC 2007 and 2012 [5] and COCO [26]. We adopt Faster R-CNN [32] as the detection method. Here we are interested in the improvements of replacing VGG-16 [41] with ResNet-101. The detection implementation (see appendix) of using both models is the same, so the gains can only be attributed to better networks.

Most remarkably, on the challenging COCO dataset we obtain a 6.0% increase in COCO’s standard metric (mAP@[.5, .95]), which is a 28% relative improvement. This gain is solely due to the learned representations.

Based on deep residual nets, we won the 1st places in several tracks in ILSVRC & COCO 2015 competitions: ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation. The details are in the appendix.

我们的方法在其他识别任务上具有良好的泛化性能。表7和表8显示了PASCAL VOC 2007和2012[5]以及COCO[26]的目标检测基线结果。我们采用Faster R-CNN[32]作为检测方法。在这里，我们对用ResNet-101取代VGG-16[41]的改进感兴趣。使用两种模型的检测实现（见附录）是相同的，因此收益只能归因于更好的网络。

最值得注意的是，在具有挑战性的COCO数据集上，我们获得了COCO标准度量（mAP@[.5，.95]）6.0%的增长，这是28%的相对改进。这种增益完全是由于学习到的表征。

基于深度残差网，我们在ILSVRC和COCO 2015比赛的几个赛道上获得了第一名：ImageNet检测、ImageNet定位、COCO检测和COCO分割。详细情况见附录。