Dense Relation Distillation with Context-aware Aggregation for Few-Shot Object Detection

Abstract

Conventional deep learning based methods for object detection require a large amount of bounding box annotations for training, which is expensive to obtain such high quality annotated data. Few-shot object detection, which learns to adapt to novel classes with only a few annotated examples, is very challenging since the fine-grained feature of novel object can be easily overlooked with only a few data available. In this work, aiming to fully exploit features of annotated novel object and capture fine-grained features of query object, we propose Dense Relation Distillation with Context-aware Aggregation (DCNet) to tackle the few-shot detection problem. Built on the meta-learning based framework, Dense Relation Distillation module targets at fully exploiting support features, where support features and query feature are densely matched, covering all spatial locations in a feed-forward fashion. The abundant usage of the guidance information endows model the capability to handle common challenges such as appearance changes and occlusions. Moreover, to better capture scale-aware features, Context-aware Aggregation module adaptively harnesses features from different scales for a more comprehensive feature representation. Extensive experiments illustrate that our proposed approach achieves state-of-the-art results on PASCAL VOC and MS COCO datasets. Code will be made available at https://github.com/hzhupku/DCNet.

传统的基于深度学习的目标检测方法需要大量的边界框注释来进行训练，而要获得这种高质量的注释数据是很昂贵的。小样本目标检测，在只有少数注释实例的情况下学习适应新的类别，是非常具有挑战性的，因为只有少数数据可用，新物体的细粒度特征很容易被忽略。在这项工作中，我们提出了带有上下文感知聚合的密集关系蒸馏法（DCNet）来解决小样本检测的问题，目的是为了充分利用注释过的新对象的特征并捕捉查询对象的细粒度特征。密集关系蒸馏模块建立在基于元学习的框架之上，旨在充分利用支持特征，支持特征与查询特征密集匹配，以前馈的方式覆盖所有空间位置。指导信息的大量使用使模型有能力处理常见的挑战，如外观变化和遮挡。此外，为了更好地捕捉尺度感知的特征，上下文感知的聚合模块自适应地利用不同尺度的特征，以获得更全面的特征表示。大量的实验表明，我们提出的方法在PASCAL VOC和MS COCO数据集上取得了最先进的结果。代码将在https://github.com/hzhupku/DCNet.

Word
a large amount of: 大量的
is expensive to: 是昂贵的
adapt to: 适应于
overlooked: v. 忽略, 俯视, 漏看, 假装不见
Built on: 建立在, 构建在, 建在, 建于
the meta-learning based framework: 基于元学习的框架
target at: 目标在
exploit: v. 开发, 利用, 剥削, 利用…谋私利 n. 功绩
in a feed-forward fashion: 以前馈的方式 以前馈方式, 前馈式的, 前馈式
abundant: adj. 大量的, 丰盛的, 充裕的
guidance information: 指导信息
endows: 赋予, 赠与, 授予, 捐赠
handle common challenges: 处理常见的挑战, 处理共同的挑战, 处理常见挑战, 处理常见的难题
Moreover: adv. 此外, 而且
Extensive experiments illustrate that: 大量实验表明
state-of-the-art: 最先进的, 最先进的技术, 先进的, 先进的技术

1. Introduction

With the success of deep convolutional neural works, object detection has made great progress these years [20, 23, 8]. The success of deep CNNs, however, heavily relies on large-scale datasets such as ImageNet [2] that enable the training of deep models. When the labeled data becomes scarce, CNNs can severely overfit and fail to generalize. While in contrast, human beings have exhibited strong performance in learning a new concept with only a few examples available. Since some object categories naturally have scarce examples or bounding box annotations are laborsome to obtain such as medical data. These problems have triggered increasing attentions to deal with learning models with limited examples. Few-shot learning aims to train models to generalize well with a few examples provided. However, most existing few-shot learning works focus on image classification [29, 26, 27] problem and only a few focus on few-shot object detection problem. Since object detection not only requires class prediction, but also demands localization of the object, making it much more difficult than few-shot classification task.
随着深度卷积神经工作的成功，这些年目标检测取得了很大的进展[20, 23, 8]. 然而，深度CNN的成功在很大程度上依赖于大规模的数据集，如ImageNet[2]，这些数据集能够对深度模型进行训练。当标记的数据变得稀少时，CNN就会严重地过度拟合，无法进行泛化。而与此相反，人类在学习一个新的概念时，在只有几个例子的情况下表现出了强大的性能。由于一些物体类别的例子很稀少，或者边界框注释的获取很费力，如医疗数据。这些问题引发了人们对处理有限实例的学习模型的日益关注。小样本学习的目的是训练模型，使其在提供少量例子的情况下就能很好地进行泛化。然而，大多数现有的小样本学习工作侧重于图像分类[29, 26, 27]问题，只有少数工作侧重于少量物体检测问题。由于物体检测不仅需要预测类别，而且还需要对物体进行定位，这使得它比几张照片的分类任务要困难得多。

Word
however: adv. 然而, 不过, 无论如何, 不管怎样 conj. 不管用什么方法, 〈古〉虽然
heavily relies on: 严重依赖
large-scale datasets: 大规模数据集
labeled data: 标记数据
scarce: adv. 勉强, 刚, 几乎不, 简直不 adj. 缺乏的, 不足的, 稀少的
fail to: v. 未能, 使…失望
generalize: v. 归纳, 概括, 笼统地讲, 概括地谈论
While in contrast: 而相反
exhibited strong performance: 表现强劲
laborsome to obtain: 很难获得
triggered : v. 触发, 扳动扳机射击, 松开扳柄 n. (枪上的)扳机, 【物】触发器, 【电】起动线路, 起动装置
These problems have triggered increasing attentions to deal with learning models with limited examples.
这些问题引发了越来越多的关注，以处理具有有限示例的学习模型。
aims to：旨在；计划做
a few examples：几个例子
focus on：专注于

Prior studies in few-shot object detection mainly consist of two groups. Most of them [13, 35, 34] adopt a meta learning [5] based framework to perform feature reweighting for a class-specific prediction. While Wang et al. [31] adopt a two-stage fine-tuning approach with only finetuning the last layer of detectors and achieve state-of-the-art performance. Wu et al. [33] also use similar strategy and focus on the scale variation problem in few-shot detection.

之前的小样本目标检测的研究主要包括两组。其中大多数[13, 35, 34]采用了基于元学习[5]的框架，对特定类别的预测进行特征重配。而Wang等人[31]采用两阶段微调方法，只对最后一层检测器进行微调，并取得最先进的性能。Wu等人[33]也采用了类似的策略，并专注于小样本检测中的尺度变化问题。

Word
Most of them：他们中的大多数
adopt ：采用 通过, 通过了, 采用了
focus on：专注于
the scale variation problem：尺度变化问题

However, aforementioned methods often suffer from several drawbacks due to the challenging nature of few shot object detection. Firstly, relations between support features and query feature are hardly fully explored in previous few-shot detection works, where global pooling operation on support features is mostly adopted to modulate the query branch, which is prone to loss of detailed local context. Specifically, appearance changes and occlusions are common for objects, as shown Fig. 1. Without enough discriminative information provided, the model is obstructed from learning critical features for class and bounding box predictions. Secondly, although scale variation problem has been widely studied in prior works [17, 15, 33], it remains a serious obstacle in few-shot detection tasks. Under few shot settings, feature extractor with scale-aware modifications is inclined to overfitting, leading to a deteriorated performance for both base and novel classes.

然而，由于小样本目标检测的挑战性性质，上述方法通常存在几个缺点。首先，在之前的小样本检测工作中几乎没有充分探索支持特征和查询特征之间的关系，其中主要采用支持特征的全局池化操作来调制查询分支，这很容易丢失详细的局部上下文。具体来说，外观变化和遮挡是 常见的物体，如图 1 所示。如果没有提供足够的判别信息，模型就会受到阻碍 从学习类和边界框预测的关键特征。其次，尽管尺度变化问题在之前的工作中得到了广泛的研究 [17, 15, 33]，但它仍然是小样本检测任务中的一个严重障碍。在小样本设置下，具有尺度感知修改的特征提取器倾向于过度拟合，导致基类和新类的性能下降。

Word
aforementioned ：adj. 前面提到的, 上述的
suffer from ：遭受
drawbacks ：缺点
due to：由于 原因是, 因为, 因
hardly ：副词: 毫不, 丝毫, 丝, 几乎不, 简直不
fully explored：充分探索
modulate ：调制，调控, 调变, 调解
is prone to：容易出现
is obstructed from：受阻于
it remains a serious obstacle：它仍然是一个严重的障碍
modifications ：n. 修改, 修正, 改进, 缓和
is inclined to：倾向于
leading to：导致

In order to alleviate the above issues, we first propose the dense relation distillation module to fully exploit support set. Given a query image and a few support images from novel classes, the shared feature learner extracts query feature and support features for subsequent matching procedure. Intuitively, the criteria that determines whether query object and support object belong to the same category mainly measures how much feature similarity they share in common. When appearance changes or occlusions occur, local detailed features are dominant for matching candidate objects and template ones. Hence, instead of obtaining global representations of support set, we propose a dense relation distillation mechanism where query and support features are matched in a pixel-wise level. Specifically, key and value maps are produced from features, which serve as encoding visual semantics for matching and containing detailed appearance information for decoding respectively. With local information of support set effectively retrieved for guidance, the performance can be significantly boosted, especially in extremely low-shot scenarios.

为了缓解上述问题，我们首先提出了密集关系蒸馏模块来充分利用支持集。给定一个查询图像和一些来自新类的支持图像，共享特征学习器提取查询 特征和支持特征用于后续匹配过程。直观地说，决定是否的标准 查询对象和支持对象属于同一类别，主要衡量它们共有多少特征相似度。当出现外观变化或遮挡时，局部细节特征对于匹配候选目标和模板目标起主导作用。因此，我们没有获得支持集的全局表示，而是提出了一种密集关系蒸馏机制，其中查询和支持特征在像素级别进行匹配。具体来说，键和值图是由特征产生的，它们分别作为用于匹配的编码视觉语义和包含用于解码的详细外观信息。通过有效检索支持集的细节信息以进行指导，可以显着提高性能，尤其是在小样本场景中。

Word
In order to：为了 为了使, 为了让, 为
alleviate ：v. 减轻, 缓解, 缓和
propose ：建议 ，提议, 提出, 提出建议
subsequent ：adj. 随后的, 后来的, 之后的, 接后的
Intuitively：直观地说，直观地讲, 直观地看, 直观上
criteria ：n. 标准, 尺度
in common：共同点
are dominant for：占主导地位
Hence：因此
mechanism ：n. 机制, 机械装置, 方法, 机件
Specifically：adv. 具体来说, 明确地, 具体地, 特意
instead of：prep. 代替, 作为…的替换
respectively：adv. 分别, 各自, 顺序为, 依次为
significantly boosted：显著提高，显著提升, 显著提高了, 显著提升了
especially ：adv. 尤其, 特别, 专门, 非常

Furthermore, for the purpose of mitigating the scale variation problem, we design the context-aware feature aggregation module to capture essential cues for different scales during RoI pooling. Since directly modifying feature extractor could result in overfitting, we choose to perform adjustment from a more flexible perspective. Recognition of objects with different scales requires different levels of contextual information, while the fixed pooling resolution may bring about loss of substantial context information. Hence, an adaptive aggregation mechanism that allocates specific attention to local and global features simultaneously could help preserve contextual information for different scales of objects. Therefore, instead of performing RoI pooling with one fixed resolution, we choose three different pooling resolutions to capture richer context features. Then an attention mechanism is introduced to adaptively aggregate output features to present a more comprehensive representation.

此外，为了减轻尺度变化问题，我们设计了上下文感知特征聚合模块，以在 RoI 池化期间捕获不同尺度的基本线索。由于直接修改特征提取器可能会导致过拟合，因此我们选择从更灵活的角度进行调整。不同尺度物体的识别需要不同层次的上下文信息，而固定的池化分辨率可能会导致大量上下文信息的丢失。因此，同时将特定注意力分配给局部和全局特征的自适应聚合机制可以帮助保留不同尺度对象的上下文信息。因此，我们选择三种不同的池化分辨率来捕获更丰富的上下文特征，而不是使用一种固定分辨率执行 RoI 池化。然后引入注意力机制来自适应地聚合输出特征以呈现更全面的表示。

Word
Furthermore：adv. 此外, 而且, 再者
mitigating ：v. 缓解, 减轻, 缓和, 平息 adj. 可考虑从轻处置的情节（或因素）
capture essential cues：捕捉重要线索
perspective：n. 观点, 远景, 景观, 透视法 adj. (按照)透视画法的, 透视的
allocates ：v. 配置, 部署, 分派, 划拨(经费等)
simultaneously ：adv. 同时, 联立, 急切地
richer ：更丰富，更为丰富, 更加丰富, 更丰富的
aggregate ：n. 骨料, 合计, 总数 v. 合计, 总计 adj. 总数的, 总计的

The contributions of this paper can be summarized as follows:

1. We propose a dense relation distillation module for few-shot detection problem, which targets at fully exploiting support information to assist the detection process for objects from novel classes.
2. We propose an adaptive context-aware feature aggregation module to better capture global and local features to alleviate scale variation problem, boosting the performance of few-shot detection.
3. Extensive experiments illustrate that our approach has achieved a consistent improvement on PASCAL VOC
   and MS COCO datasets. Specially, our approach achieves better performance than the state-of-the-art methods on the two datasets.

本文的贡献可以总结如下：

1. 我们为小样本检测问题提出了一个密集关系蒸馏模块，其目标是充分利用支持信息来辅助新类别对象的检测过程。
2. 我们提出了一个自适应上下文感知特征聚合模块，以更好地捕捉全局和局部特征，以缓解尺度变化问题，提高小样本检测的性能。
3. 大量实验表明，我们的方法在 PASCAL VOC 和 MS COCO 数据集上取得了一致的改进。特别是，我们的方法在两个数据集上实现了比最先进的方法更好的性能。

Word
targets at：目标在
assist ：v. 协助, 帮助, 援助, 促进 n. （曲棍球等）助攻, （棒球等）助杀
alleviate ：v. 减轻, 缓解, 缓和
Specially：adv. 特别, 特意, 尤其, 专门地
our approach achieves better performance than the state-of-the-art methods on the two datasets.
我们的方法在两个数据集上实现了比最先进的方法更好的性能。

2.相关工作

2.1 一般的目标检测

Deep learning based object detection can be mainly divided into two categories: one-stage and two-stage detectors. One-stage detector YOLO series [20, 21, 22] provide a proposal-free framework, which uses a single convolutional network to directly perform class and bounding box predictions. SSD [18] uses default boxes to adjust to various object shapes. On the other hand, RCNN and its variants [7, 9, 6, 23, 8] fall into the second category. These methods first extract class-agnostic region proposals of the potential objects from a given image. The generated boxes are then further refined and classified into different categories by subsequent modules. Moreover, many works are proposed to handle scale variance [17, 15, 24, 25]. Compared to one-stage methods, two-stage methods are slower but exhibit better performance. In our work, we adopt Faster RCNN as the base detector.

基于深度学习的目标检测主要可以分为两类：单阶段检测器和两阶段检测器。单阶段检测器 YOLO 系列 [20, 21, 22] 提供了一个无提议框架，它使用单个卷积网络直接执行类别和边界框预测。 SSD [18] 使用默认框来调整各种目标形状。另一方面，RCNN 及其变体 [7, 9, 6, 23, 8] 属于第二类。这些方法首先从给定图像中提取潜在目标的类别不可知区域提议。然后生成的框被后续模块进一步细化并分类为不同的类别。此外，提出了许多工作来处理尺度方差 [17, 15, 24, 25]。与单阶段方法相比，两阶段方法速度较慢，但​​表现出更好的性能。在我们的工作中，我们采用 Faster RCNN 作为基础检测器。

Word
are proposed to：拟将，建议将, 建议, 拟在

2.2 小样本学习

Few-shot learning aims to learn transferable knowledge that can be generalized to new classes with scarce examples. Bayesian inference is utilized in [4] to generalize knowledge from a pretrained model to perform one-shot learning. Meta-learning based methods have been prevalent in few-shot learning these days. Metric learning based methods [16, 29, 26, 27] have achieved state-of-the-art performance in few-shot classification tasks. Matching Network [29] encodes input into deep neural features and performs weighted nearest neighbor matching to classify query images. Our proposed method is also based on matching mechanism. Prototypical Network [26] represents each class with one prototype which is a feature vector. Relation Network [27] learns a distance metric to compare the target image with a few labeled images. While optimization based methods [19, 5] are proposed for fast adaptation to new few-shot task. [11] proposes a cross-attention mechanism to learn correlations between support and query images. Above methods are focusing on the few-shot classification task while few-shot object detection problem is relatively under-explored.

小样本学习旨在学习可迁移的知识，这些知识可以推广到具有稀缺示例的新类。 [4] 中利用贝叶斯推理从预训练模型中概括知识以执行一次性学习。如今，基于元学习的方法在小样本学习中很流行。基于度量学习的方法 [16, 29, 26, 27] 在少样本分类任务中取得了最先进的性能。匹配网络 [29] 将输入编码为深度神经特征，并执行加权最近邻匹配以对查询图像进行分类。我们提出的方法也是基于匹配机制。原型网络 [26] 用一个原型表示每个类，原型是一个特征向量。关系网络 [27] 学习距离度量以将目标图像与一些标记图像进行比较。虽然提出了基于优化的方法 [19, 5] 以快速适应新的少拍任务。 [11] 提出了一种交叉注意机制来学习支持和查询图像之间的相关性。上述方法侧重于少样本分类任务，而少样本对象检测问题相对未得到充分探索。

Word
aims to：旨在
inference ：名词: 推理, 推论, 论断, 意味
is utilized in：被用于
under-explored：未充分开发的

2.3 小样本目标检测

Few-shot object detection aims to detect object from novel classes with only a few annotated training examples provided. LSTD [1] and RepMet [14] adopt a general transfer learning framework which reduces overfitting by adapting pre-trained detectors to few-shot scenarios. Recently, Meta YOLO [13] designs a novel few-shot detection model with YOLO v2 [21] that learns generalizable meta features and automatically reweights the features for novel classes by producing class-specific activating coefficients from support examples. Meta R-CNN [35] and FsDetView [34] perform similar process with base detector as Faster R-CNN. TFA [31] simply performs two-stage finetuning approach by only finetuning the classifier on the second stage and achieves better performance. MPSR [33] proposes multiscale positive sample refinement to handle scale variance problem. CoAE [12] proposes non-local RPN and focuses on one-shot detection from the view of tracking by comparing itself with other tracking methods, while our method performs cross-attention on features extracted by the backbone in a more traightforward way and targets at few-shot detection task. FSOD [3] proposes attention-RPN, multi-relation detector and contrastive training strategy to detect novel object. In our work, we adopt the similar meta-learning based framework as Meta R-CNN and further improve the performance. Moreover, with our proposed method, the class-specific prediction procedure can be successfully removed, simplifying the overall process.

小样本目标检测旨在仅使用少数带注释的训练示例从新类别中检测目标。假如。 LSTD [1] 和 RepMet [14] 采用通用的迁移学习框架，通过将预训练的检测器适应小样本场景来减少过度拟合。最近，Meta YOLO [13] 使用 YOLO v2 [21] 设计了一种新颖的小样本检测模型，该模型学习可泛化的元特征，并通过从支持示例中生成特定于类的激活系数来自动重新加权新类的特征。 Meta R-CNN [35] 和 FsDetView [34] 使用基础检测器执行与 Faster R-CNN 类似的过程。 TFA [31] 通过仅在第二阶段微调分类器来简单地执行两阶段微调方法，并获得更好的性能。 MPSR [33] 提出了多尺度正样本细化来处理尺度方差问题。 CoAE [12] 提出了 non-local RPN 并通过与其他跟踪方法进行比较，从跟踪的角度专注于单次检测，而我们的方法以更直接的方式对主干提取的特征进行交叉注意，并针对小样本检测任务。 FSOD [3] 提出了注意力 RPN、多关系检测器和对比训练策略来检测新对象。在我们的工作中，我们采用了与 Meta R-CNN 类似的基于元学习的框架，并进一步提高了性能。此外，使用我们提出的方法，可以成功删除特定于类的预测过程，从而简化整个过程。

Word
automatically ：adv. 自动地, 自然地, 无意识地, 不自觉地
achieves ：v. 实现, 做到, 获得(胜利等), 取得预期效果

3. 方法

3.1 准备工作

Problem Definition. Following setting in [13, 35], object classes are divided into base classes $C_{base}$ with abundant annotated data and novel classes $C_{novel}$ with only a few annotated samples, where $C_{base}$ and $C_{novel}$ have no intersection. We aim to obtain a few-shot detection model with the ability to detect objects from both base and novel classes in testing by leveraging generalizable knowledge from base classes. The number of instances per category for novel classes is set as $k$ ($i.e.,k-shot$).

问题定义。 按照 [13, 35] 中的设置，目标类被分为具有丰富注释数据的基类 $C_{base}$ 和只有少量注释样本的新类 $C_{novel}$，其中 $C_{base}$ 和 $C_{novel}$ 没有交集。我们的目标是通过利用基类的泛化知识，获得一个能够在测试中从基类和新类中检测对象的小样本检测模型。新类的每个类别的实例数设置为 $k$（$i.e.,k-shot$）。

Word
are divided into：被分成
abundant annotated data：丰富的注释数据
a few annotated samples： 几个带注释的样本
intersection：n. 交叉, 相交, 十字路口, 交叉路口 交集
aim to：目标是
leveraging ：利用
We aim to obtain a few-shot detection model with the ability to detect objects from both base and novel classes in testing by leveraging generalizable knowledge from base classes.
我们的目标是通过利用基类的泛化知识，获得一个能够在测试中从基类和新类中检测对象的小样本检测模型。

We align the training scheme with the episodic paradigm [29] for few-shot scenario. Given a $k-shot$ learning task, each episode is constructed by sampling: 1) a support set containing image-mask pairs for different classes S = { x i , y i } i = 1 N S = \left\{ {{x_i},{y_i}} \right\}_{i = 1}^N S={xi​,yi​}i=1N​ ,where $x_i\in {\mathbb{R}}^{h\times w\times 3}$ is an RGB image, $y_i\in {\mathbb{R}}^{h\times w}$is a binary mask for objects of class $i$ in the support image generated from bounding box annotations and $N$ is the number of classes in the training set; 2) a query image $q$ and annotations $m$ for the training classes in the query image. The input to the model is the support pairs and query image, the output is detection prediction for query image.

我们将训练方案与小样本场景的情景范例 [29] 保持一致。给定一个 $k-shot$ 学习任务，每个情节都是通过采样构建的：1) 包含不同类别的图像掩码对的支持集 S = { x i , y i } i = 1 N S = \left\{ {{x_i},{y_i}} \right\} _{i = 1}^N S={xi​,yi​}i=1N​ ，其中 $x_i\in {\mathbb{R}}^{h\times w\times 3}$ 是一个 RGB 图像， $y_i\in {\mathbb{R}}^{h\times w}$ 是从边界框注释生成的支持图像中 $i$ 类对象的二进制掩码，$N$ 是训练集中的类数； 2）查询图像中的训练类的查询图像 $q$ 和注释 $m$。模型的输入是支持对和查询图像，输出是查询图像的检测预测。

Word
the episodic paradigm：事件范式
is constructed by：构建的是, 构建了, 构建的

Basic Object Detection. The choice of base detectors is varied. [13] utlizes YOLO v2 [21] which is a one-stage detector, while [35] adopts Faster R-CNN [23] which is a two-stage detector and provides consistently better results.Therefore, we also adopt Faster R-CNN as our base detector which consists of a feature extractor, region proposal network (RPN) and the detection head (RoI head).

基本目标检测。基检测器的选择是多样的。[13]使用YOLO v2 [21]，这是一个单阶段的检测器，而[35]采用Faster R-CNN [23]，这是一个两阶段的检测器，并始终提供更好的结果。因此，我们也采用Faster R-CNN作为我们的基础检测器，它由一个特征提取器、区域建议网络（RPN）和检测头（RoI头）组成。

Word
The choice of base detectors is varied.：基检测器的选择是多样的。
utlizes ：利用
adopts ：采用
provides：提供
Therefore：adv. 因此

Feature Reweighting for Detection. We choose Meta-RCNN [35] as our baseline method. Formally, let $I$ denote an input query image, { I s i , M s i } ∣ i = 1 N \left\{ {{I_{si}},{M_{si}}} \right\}\left| {_{i = 1}^N} \right. {Isi​,Msi​}∣∣​i=1N​ denote support images and masks converted from bounding-box annotations, where $N$ is the number of training classes. RoI features $Z^j|{}_{j=1}^n$ is generated by the RoI pooling layer ($n$ is the number of RoIs) and class-specific vectors $w_i\in {\mathbb{R}}^C,i=1,2,...,N$ are produced with a reweighting module which shares its backbone parameters with the feature extractor, where $C$ is the feature dimension. Then class-specific feature $z_i$ is achieved with: $$z_i=z\otimes w_i,i==1,2,...,N$$ where $\otimes$ denotes channel-wise multiplication. Then class-specific prediction is performed to output the detection results. Based on this methodology, we further make a significant improvement and simplify the prediction procedure by removing the class-specific prediction.

用于检测的特征重新加权。我们选择 Meta-RCNN [35] 作为我们的基线方法。形式上，让 $I$ 表示输入查询图像， { I s i , M s i } ∣ i = 1 N \left\{ {{I_{si}},{M_{si}}} \right\}\left| {_{i = 1}^N} \right. {Isi​,Msi​}∣∣​i=1N​ 表示从边界框注释转换的支持图像和掩码，其中 $N$ 是训练类的数量。投资回报率特征 $Z^j|{}_{j=1}^n$ 由 RoI 池化层生成（$n$ 是 RoI 的数量）和特定于类的向量 $w_i\in {\mathbb{R}}^C,i=1,2,...,N$ 由重加权模块生成，该模块与特征提取器共享其主干参数，其中 $C$ 是特征维度。然后类特定的特征 $z_i$ 实现为： $$z_i=z\otimes w_i,i==1,2,...,N$$ 其中 $\otimes$ 表示通道层次乘法。然后进行类别特定的预测以输出检测结果。基于这种方法，我们通过删除特定于类的预测进一步进行了重大改进并简化了预测过程。

Word
Formally：adv. 形式上, 正式地, 遵照一定格式地
class-specific：特定于类
Based on ：基于
simplify ：v. 使简化, 使简易

3.2 DCNet

As illustrated in Fig. 2, we present the Dense Relation Distillation (DRD) module with Context-aware Feature Aggregation (CFA) module to fully exploit support features and capture essential context information. The two proposed components form the final model DCNet. We will first depict the architecture of the proposed DRD module. Then we will bring out the details of the CFA module.

如图 2 所示，我们展示了具有上下文感知特征聚合 (CFA) 模块的密集关系蒸馏 (DRD) 模块，以充分利用支持特征并捕获必要的上下文信息。这两个提议的组件构成了最终模型 DCNet。我们将首先描述所提出的 DRD 模块的架构。然后我们将带出CFA模块的细节。

Word
As illustrated in Fig. 2：如图 2 所示
present ：n. 目前, 现在, 礼物, 礼品 adj. 存在, 出席, 在场, 出现 v. 出现, 提出, 显示, 提交
fully exploit ：充分利用
form ：动词: 形成, 构成, 构, 结, 组织, 组, 树立, 缔, 发展, 养 。名词: 形式, 表格, 形态, 格式, 形, 形状, 式, 体, 态, 样式, 结成, 程式, 形像, 单子
bring out ：na. 公布, 说出, 出版, 上演

3.2.1 Dense Relation Distillation Module 密集关系蒸馏模块

Key and Value Embedding. Given a query image and support set, query and support features are produced by feeding them into the shared feature extractor. The input of the dense relation distillation (DRD) module is the query feature and support features. Both parts are first encoded into pairs of key and value maps through the dedicated deep encoders. The query encoder and support encoder adopt the same structure while not sharing parameters.

键和值嵌入。 给定查询图像和支持集，通过将它们输入共享特征提取器来生成查询和支持特征。密集关系蒸馏（DRD）模块的输入是查询特征和支持特征。这两个部分首先通过专用的深度编码器编码成键和值映射对。查询编码器和支持编码器采用相同的结构，但不共享参数。

Word
are produced by：是由 是由以下方面产生的,
feeding them into：将它们送入，把它们送入, 把它们送进, 把它们喂给
Both parts are first encoded into：两部分首先被编码成
adopt ：动词: 采用, 采取, 通过, 采纳, 收养, 抱, 取, 树立, 认, 义

The encoder takes one or multiple feature as input and outputs two feature maps for each input feature: key and value with two parallel $3\times 3$ convolution layers, which serve as reducing the dimension of the input feature to save computation cost. Specifically, key maps are used for measuring the similarities between query feature and support features, which help determine where to retrieve relevant support values. Therefore, key maps are learned to encode visual semantics for matching and value maps store detailed information for recognition. Hence, for query feature, the output is a pair of key and value maps:$k_q\in {\mathbb{R}}^{C/8\times H\times W}$ , $v_q\in {\mathbb{R}}^{C/2\times H\times W}$ , where $C$ is the feature dimension, $H$ is the height, and $W$ is the width of input feature map. For support features, each of the features is independently encoded into key and value maps, the output is $k_s\in {\mathbb{R}}^{N\times C/8\times H\times W}$ , $k_s\in {\mathbb{R}}^{N\times C/8\times H\times W}$ , where $N$ is the number of target classes (also the number of support samples). The generated key and value maps are further fed into the relation distillation part where keys maps of query and support are densely matched for addressing target objects.

编码器将一个或多个特征作为输入，并为每个输入特征输出两个特征映射：key 和 value，具有两个并行的 $3\times 3$ 卷积层，用于降低输入特征的维度以节省计算成本。具体来说，键映射用于衡量查询特征和支持特征之间的相似性，这有助于确定在哪里检索相关支持值。因此，学习关键图来编码视觉语义以进行匹配，而值图存储用于识别的详细信息。因此，对于查询特征，输出是一对键值映射：$k_q\in {\mathbb{R}}^{C/8\times H\times W}$ , $v_q\in {\mathbb{R}}^{C/2\times H\times W}$ ，其中 $C$ 是特征维度，$H$ 是高度，$W$ 是输入特征图的宽度。对于支持功能，每个功能都独立编码为键和值映射，输出为 $k_s\in {\mathbb{R}}^{N\times C/8\times H\times W}$ ， $k_s\in {\mathbb{R}}^{N\times C/8\times H\times W}$ ，其中 $N$ 是目标类的数量（也是支持样本的数量）。生成的键和值映射被进一步输入到关系蒸馏部分，其中查询和支持的键映射被密集匹配以寻址目标对象。

Word
parallel :平行 ,平行的, 平行线, 纬度
save computation cost:节省计算成本, 节省计算费用, 节约计算成本, 节省了计算成本
Specifically:adv. 具体来说, 明确地, 具体地, 特意
determine :确定, 决定, 判断, 决定了
retrieve :v. 找回, 取回, 挽回, 索回 n. 恢复
Therefore:adv. 因此
Hence:adv. 因此, 由此 int. 〈诗〉去
independently :adv. 独立地, 自由地
fed into:un. 注入

Relation Distillation. After acquiring the key/value maps of query and support features, relation distillation is performed. As illustrated in Fig. 2, soft weights for value maps of support features are computed via measuring the similarities between key maps of query feature and support features. The pixel-wise similarity is performed in a non-local manner, formulated as: $$F(k_{qi},k_{sj})=\varphi (k_{qi}{)}^T{\varphi }^{\prime }(k_{sj}),$$ where $i$ and $j$ are the index of the query and support location, $\varphi$, ${\varphi }^{\prime }$ denote two different linear transformations with parameters learned via back propagation during training process, forming a dynamically learned similarity function. After computing the similarity of pixel features, we perform softmax normalization to output the final weight $W$ : $$W_{ij}=\frac{\exp (F(k_{qi},k_{sj}))}{\sum \exp (F(k_{qi},k_{sj}))}.$$ Then the value of the support features are retrieved by a weighted summation with the soft weights produced and then it is concatenated with the value map of query feature. Hence, the final output is formulated as: $$y=concat[v_q,W\ast v_s],$$ where $\ast$ denotes matrix inner-product. Noted that there are $N$ support features, which brings $N$ key-value pairs. We perform summation over $N$ output results to obtain the final result, which is a refined query feature, activated by support features where there are co-existing classes of objects in query and support images.

关系蒸馏。 在获取查询和支持特征的键/值图后，进行关系蒸馏。如图 2 所示，支持特征值图的软权重是通过测量查询特征的关键映射和支持特征之间的相似性来计算的。以非局部方式执行像素级相似性，公式如下：$$F(k_{qi},k_{sj})=\varphi (k_{qi}{)}^T{\varphi }^{\prime }(k_{sj}),$$ 其中$i$和$j$是查询和支持位置的索引，$\varphi$、${\varphi }^{\prime }$表示两个不同的线性变换，参数在训练过程中通过反向传播学习，形成一个动态学习的相似度函数。在计算像素特征的相似度后，我们执行 softmax 归一化以输出最终权重 $W$：$$W_{ij}=\frac{\exp (F(k_{qi},k_{sj}))}{\sum \exp (F(k_{qi},k_{sj}))}.$$ 然后通过加权求和与产生的软权重来检索支持特征的值，然后将其与查询特征的值图连接起来。因此，最终输出公式为：$$y=concat[v_q,W\ast v_s],$$ 其中 $\ast$ 表示矩阵内积。注意到有 $N$ 支持特征，它带来了 $N$ 个 键值对。我们对 $N$ 输出结果进行求和以获得最终结果，这是一个细化的查询特征，由支持特征激活，其中查询和支持图像中的目标类共存。

Figure 2. The overall framework of our proposed DCNet. For training, the input for each episode consists of a query image and N support image-mask pairs from N classes. The shared feature extractor first produces query feature and support features. Then, the dense relation distillation (DRD) module performs dense feature match to activate co-exisiting features of input query. With proposals produced by RPN, context-aware feature aggregation (CFA) module adaptively harnesses features generated with different scales of pooling operations, capturing different levels of features for a more comprehensive representation.

图 2. 我们提出的 DCNet 的整体框架。对于训练，每一批次的输入包括一个查询图像和来自 N 个类别的 N 个支持图像-掩码对。共享特征提取器首先生成查询特征和支持特征。然后，密集关系蒸馏（DRD）模块执行密集特征匹配以激活输入查询的共存特征。通过 RPN 生成的提议，上下文感知特征聚合 (CFA) 模块自适应地利用不同规模的池化操作生成的特征，捕获不同级别的特征以获得更全面的表示。

Word
acquiring ：获取
is performed： 被执行
features are computed via measuring the similarities：通过测量相似性来计算特征
denote ：动词: 表示, 意味着
are retrieved by：被检索
Hence：因此
is formulated as：被公式化为
produces ：产生

Previous trials [13, 35, 34] utilize class-wise vectors generated by global pooling of support features to modulate the query feature, which guide the feature learning from a holistic view. However, since appearance changes or occlusions are common in natural images, the holistic feature may be misleading when objects of the same class vary much between query and support samples. Also, when most parts of the objects are unseen due to the occlusions, the retrieval of local detailed features becomes substantial, which former methods completely neglect. Hence, equipped with the dense relation distillation module, pixel-level relevant information can be distilled from support features. As long as there exist some common characteristics, the pixels of query features belonging to the co-existing objects between query and support samples will be further activated, providing a robust modulated feature to facilitate the prediction of class and bounding-box.

先前的试验 [13, 35, 34] 利用支持特征的全局池化生成的类向量来调制查询特征，从整体角度指导特征学习。然而，由于外观变化或遮挡在自然图像中很常见，当同一类的对象在查询和支持样本之间差异很大时，整体特征可能会产生误导。此外，当物体的大部分由于遮挡而看不见时，局部细节特征的检索变得很重要，而以前的方法完全忽略了这一点。因此，配备了密集关系蒸馏模块，可以从支持特征中提取像素级相关信息。只要存在一些共同特征，属于查询样本和支持样本之间共存对象的查询特征的像素将被进一步激活，提供一个鲁棒的调制特征，以促进类和边界框的预测。

Word
vectors generated by：生成的向量
guide ：名: 指南, 导轨, 先导, 讲解员, 向, 路标 动词: 指导, 引导, 导向, 导, 向导, 指引, 导引, 带领, 引, 向, 南针, 羑
a holistic view：整体观
However：然而
holistic ：整体的
Also：adv. 也, 同样, 此外, 而且 conj. 同“also beautiful”
due to ：介词: 由于, 由, 基于, 承, 定于
retrieval ：名词: 恢复, 取回
substantial：adj. 大量的, 价值巨大的, 重大的, 大而坚固的
Hence：因此
equipped with：配备
As long as ：只要
modulated ：调制的

Our distillation method can be seen as an extension of the non-local self-attention mechanism [28, 30]. However, instead of performing self-attention, we specially design the relation distillation model to realize information retrieval from support features to modulate the query feature, which can be treated as a cross attention.

我们的蒸馏方法可以看作是非局部自注意力机制的扩展 [28, 30]。然而，我们并没有执行自注意力，而是专门设计了关系蒸馏模型来实现从支持特征到调制查询特征的信息检索，可以将其视为交叉注意力。

Word
instead of ：介词: 顶, 而不是
be treated as：把……当作……对待；当做……来看；被当作

3.2.2 Context-aware Feature Aggregation 上下文感知特征聚合

After performing dense relation distillation, DRD module has fulfilled its duty. The refined query feature is subsequently fed into RPN where region proposals are output. Taking proposals and feature as input, RoI Align module performs feature extraction for final class prediction and bounding-box regression. Normally, pooling operation is implemented with a fixed resolution 8 in our original implementation, which is likely to cause information loss during training. For general object detection, this kind of information loss can be remedied with large scale of training data, while the problem becomes severe in few-shot detection scenarios with only a few training data available, which is inclined to induce a misleading detection results. Moreover, with scale variation amplified due to the few-shot nature, the model tends to lose the generalization ability to novel classes with adequate adaption to different scales. To this end, we propose Context-aware Feature Aggregation (CFA) module. Instead of using a fixed resolution 8, we empirically choose 4, 8 and 12 three resolutions and perform parallel pooling operation to obtain a more comprehensive feature representation. The larger resolution tends to focus on local detailed context information specially for smaller objects, while the smaller resolution targets at capturing holistic information to benefit the recognition of larger objects, providing a simple and flexible way to alleviate the scale variation problem.

在进行了密集关系蒸馏之后，DRD 模块就完成了它的任务。细化的查询特征随后被输入到 RPN 中，在那里输出区域建议。以提议和特征为输入，RoI Align 模块为最终类别预测和边界框回归执行特征提取。通常情况下，池化操作在我们最初的实现中是用固定的分辨率8来实现的，这很可能导致训练过程中的信息丢失。对于一般的目标检测，这种信息丢失可以通过大规模的训练数据来弥补，而在只有少量训练数据可用的小样本检测场景中，问题变得严重，容易导致误导性的检测结果。此外，由于小样本性质导致尺度变化被放大，该模型往往会失去对新类的泛化能力，但对不同尺度有足够的适应能力。为此，我们提出了上下文感知特征聚合（CFA）模块。我们没有使用固定的分辨率 8，而是凭经验选择 4、8 和 12 三个分辨率并进行并行池化操作以获得更全面的特征表示。较大的分辨率倾向于专注于较小物体的局部详细上下文信息，而较小的分辨率则侧重于捕获整体信息以有利于较大物体的识别，为缓解尺度变化问题提供了一种简单灵活的方法。

Word
module has fulfilled its duty：模块已经完成任务
subsequently ：副词: 后来, 以后, 既而
fed into ：送入 输送到, 被送入, 送到
feature is subsequently fed into RPN where region proposals are output. ：特征随后被输入到 RPN 中，在那里输出区域提议。
Taking proposals and feature as input：将提案和特征作为输入
Normally：adv. 通常, 正常地, 正常情况下, 平常地
be remedied with：补救
severe ：adj. 极为恶劣的, 十分严重的, 严厉的, 苛刻的
is inclined to：倾向于
Moreover：而且
with scale variation amplified：放大了尺度变化
tends to：倾向于
adequate ：形容词: 充足, 足够, 适当, 够
To this end, ：un. 直到最后, 为此目的
propose ：提供
empirically ：凭经验
targets at：目标在
alleviate ：动词: 缓和, 和缓

Since each generated feature contains different level of semantic information. With the intention to efficiently aggregate features generated from different scales of RoI pooling, we further propose an attention mechanism to adaptively fuse the pooling results. As illustrated in Fig. 3, we add an attention branch for each feature which consists of two blocks. The first block contains a global average pooling. The second one contains two consecutive fc layers. Afterwards, we add a softmax normalization to the generated weights for balancing the contribution of each feature. Then the final output of the aggregated feature is the weighted summation of the three features.

由于每个生成的特征包含不同级别的语义信息。为了有效地聚合从不同规模的 RoI 池化生成的特征，我们进一步提出了一种注意力机制来自适应地融合池化结果。如图 3 所示，我们为每个由两个块组成的特征添加一个注意力分支。第一个块包含全局平均池化。第二个包含两个连续的 fc 层。之后，我们向生成的权重添加 softmax 归一化，以平衡每个特征的贡献。那么聚合特征的最终输出就是三个特征的加权求和。

Word
aggregate ：融合
fuse：融合
consecutive ：adj. 连续不断的
contribution ：n. 贡献, 捐款, 捐赠, 稿件
Afterwards：adv. 后来, 以后

Figure 3. Illustration of context-aware feature aggregation. Attention mechanism is adopted to adaptively aggregate different features, where the weights are normalized with softmax function.
图 3. 上下文感知特征聚合的图示。采用注意力机制自适应聚合不同的特征，其中权重使用 softmax 函数进行归一化。

3.3. Learning Strategy 学习策略

As illustrated in Fig. 4, we follow the training paradigm in [13, 35, 34], which consists of meta-training and meta fine-tuning. In the phase of meta-training, abundant annotated data from base classes is provided. We jointly train the feature extractor, dense relation distillation module, context-aware feature aggregation module and other basic components of detection model. In meta fine-tuning phase, we train the model on both base and novel classes. As only $k$ labeled bounding-boxes are available for the novel classes, to balance between samples from base and novel classes, we also include $k$ boxes for each base class. The training procedure is the same as the meta-training phase but with fewer iterations for model to converge.

如图 4 所示，我们遵循 [13, 35, 34] 中的训练范式，其中包括元训练和元微调。在元训练阶段，提供了大量来自基类的注释数据。我们联合训练特征提取器、密集关系蒸馏模块、上下文感知特征聚合模块和检测模型的其他基本组件。在元微调阶段，我们在基类和新类上训练模型。由于只有 $k$ 个标记的边界框可用于新类，为了平衡来自基类和新类的样本，我们还为每个基类包含了 $k$ 个框。训练过程与元训练阶段相同，但模型收敛的迭代次数更少。

Word
jointly ：副词: 共同
components ：成分
fewer iterations:更少的迭代
converge:v. 集中, 汇集, 聚集, （向某一点）相交

Figure 4. Demonstration of learning strategy of meta-learning based few-shot detection framework. The meta learner aims to acquire meta information and help the model to generalize to novel classes.
图 4. 基于元学习的少样本检测框架的学习策略演示。元学习器旨在获取元信息并帮助模型推广到新的类。

4. Experiments 实验

In this section, we first introduce the implementation details and experimental configurations in Sec. 4.1. Then we present our detailed experimental analysis on PASCAL VOC dataset in Sec. 4.2 together with ablation studies and qualitative results. Finally, results on COCO dataset will be presented in Sec. 4.3.

在本节中，我们首先介绍第二节中的实现细节和实验配置。 4.1.然后我们在第二节中介绍了我们对 PASCAL VOC 数据集的详细实验分析。 4.2 连同消融研究和定性结果。最后，在 COCO 数据集上的结果将在 Sec. 4.3.

Word
In this section, :在这个部分，
configurations:n. 配置, 结构, 外形, 组合
Finally:adv. 终于, 最终, （用于列举）最后, 彻底地

4.1. Datasets and Settings 数据集和设置

Following the instructions in [13], we construct the few-shot detection datasets for fair comparison with other state-of-the-art methods. Moreover, to achieve a more stable few-shot detection results, we perform 10 random runs with different randomly sampled shots. Hence, all the results in theexperiments is averaged results by 10 random runs.

按照 [13] 中的说明，我们构建了小样本检测数据集，以便与其他最先进的方法进行公平比较。此外，为了获得更稳定的少镜头检测结果，我们使用不同的随机采样镜头执行 10 次随机运行。因此，实验中的所有结果都是 10 次随机运行的平均结果。

Word
instructions ： n. 指令, 教导, 教训
Moreover：adv. 此外, 而且
is averaged results：是平均结果

PASCAL VOC. For PASCAL VOC dataset, we train our model on the VOC 2007 trainval and VOC 2012 trainval sets and test the model on VOC 2007 test set. The evaluation metric is the mean Average Precision (mAP). Both the train-val sets are split by object categories, where 5 are randomly chosen as novel classes and the left 15 are base classes. We use the same split as [13], where novel classes for four splits are {“bird”, “bus”, “cow”, “motorbike” (“mbike”), “sofa”}, {“aeroplane”(“aero”, “bottle”, “cow”, “horse”, “sofa”}, {“boat”, “cat”, “motorbike”, “sheep”, “sofa”}, respectively. For few-shot object detection experiments, the few-shot dataset consists of images where $k$ object instances are available for each category and $k$ is set as 1/3/5/10. COCO. MS COCO dataset has 80 object categories, where the 20 categories overlapped with PASCAL VOC are set to be novel classes. 5000 images from the validation set noted as minival are used for evaluation while the left images in the train and validation set are used for training. The process of constructing few-shot dataset is similar to PASCAL VOC dataset and k is set as 10/30.

PASCAL VOC. 对于PASCAL VOC数据集，我们在VOC 2007 trainval和VOC 2012 trainval集上训练我们的模型，并在VOC 2007测试集上测试该模型。评估指标是平均平均精度（mAP）。这两个训练集和测试集都是按对象类别划分的，其中5个是随机选择的新类别，其余15个是基本类别。我们使用与[13]相同的拆分，其中四个拆分的新颖类分别是{“鸟”、“公共汽车”、“牛”、“摩托车”（“mbike”）、“沙发”}、{“飞机”（“aero”、“瓶子”、“牛”、“马”、“沙发”}、{“船”、“猫”、“摩托车”、“羊”、“沙发”}。对于小样本的物体检测实验，几张照片的数据集由图像组成，每个类别都有$k$个物体实例，$k$被设定为1/3/5/10。COCO。MS COCO数据集有80个物体类别，其中与PASCAL VOC重叠的20个类别被设定为新的类别。来自验证集的5000张被注为minival的图像被用于评估，而训练集和验证集的左侧图像被用于训练。构建小样本数据集的过程与PASCAL VOC数据集相似，K被设定为10/30。

Word
evaluation metric：评价指标
noted as: 以…而闻名
minival ：迷你

Implementation Details. We perform training and testing process on images with a single scale. The shorter side of the query image is resized to 800 pixels and longer sides are less than 1333 pixels while maintaining the aspect ratio. The support image is resized to a squared image of 256 × 256. We adopt ResNet-101 [10] as feature extractor and RoI Align [8] as RoI feature extractor. The weights of the backbone is pre-trained on ImageNet [2]. After training on base classes, only the last fully-connected layer (for classification) is removed and replaced by a new one randomly initialized. It is worth noting that all parts of the model participate in learning process in the second meta fine-tuning phase without any freeze operation. We train our model with a mini-batch size as 4 with 2 GPUs. We utilize the SGD optimizer with the momentum of 0.9, and weight decay of 0.0001. For meta-training on PASCAL VOC, models are trained for 240k, 8k, and 4k iterations with learning rates of 0.005, 0.0005 and 0.00005 respectively. For meta fine-tuning on PASCAL VOC, models are trained for 1300, 400 and 300 iterations with learning rates as 0.005, 0.0005 and 0.00005 respectively. As for MS COCO dataset, during meta-training, models are trained for 56k, 14k and 10k iterations with learning rates of 0.005, 0.0005 and 0.00005 respectively. And during meta fine-tuning, model are trained for 2800, 700 and 500 iteration for 10-shot fine-tuning and 5600, 1400 and 1000 iterations for 30-shot fine-tuning. Baseline Method. Since we adopt Faster-RCNN as base detector, we choose Meta R-CNN [35] as the baseline method. Moreover, we implement it by ourselves for a more fair comparison.

实施细节。 我们在单一尺度的图像上执行训练和测试过程。查询图像的短边调整为 800 像素，长边小于 1333 像素，同时保持纵横比。支持图像被调整为 256 × 256 的平方图像。我们采用 ResNet-101 [10] 作为特征提取器，使用 RoI Align [8] 作为 RoI 特征提取器。主干的权重在 ImageNet [2] 上进行了预训练。在对基类进行训练后，只有最后一个全连接层（用于分类）被移除并替换为一个随机初始化的新层。值得注意的是，模型的所有部分都参与了第二个元微调阶段的学习过程，没有任何冻结操作。我们使用 2 个 GPU 以 4 的 mini-batch 大小训练我们的模型。我们使用动量为 0.9，权重衰减为 0.0001 的 SGD 优化器。对于 PASCAL VOC 的元训练，模型训练 240k、8k 和 4k 次迭代，学习率分别为 0.005、0.0005 和 0.00005。对于 PASCAL VOC 的元微调，模型训练 1300、400 和 300 次迭代，学习率分别为 0.005、0.0005 和 0.00005。对于 MS COCO 数据集，在元训练期间，模型训练 56k、14k 和 10k 次迭代，学习率分别为 0.005、0.0005 和 0.00005。在元微调期间，模型训练 2800、700 和 500 次迭代进行 10 次微调，5600、1400 和 1000 次迭代进行 30 次微调。基线方法。由于我们采用 Faster-RCNN 作为基础检测器，因此我们选择 Meta R-CNN [35] 作为基线方法。此外，我们自己实施它以进行更公平的比较。

Word
the aspect ratio：纵横比
all parts of：的所有部分
participate ：v. 参与, 参加

4.2 Experiments on PASCAL VOC PASCAL VOC实验

In this section, we conduct experiments on PASCAL VOC dataset. We first compare our method with the state-of-the-art methods. Then we carry out ablation studies to perform comprehensive analysis of the components of our proposed DCNet. Finally, some qualitative results are presented to provide an intuitive view of the validity of our method. For all the experiments, we run 10 trials with random support data and report the averaged performance.

在本节中，我们在 PASCAL VOC 数据集上进行实验。我们首先将我们的方法与最先进的方法进行比较。然后我们进行消融研究，对我们提出的 DCNet 的组件进行综合分析。最后，提出了一些定性结果，以提供我们方法有效性的直观视图。对于所有实验，我们使用随机支持数据运行 10 次试验并报告平均性能。

Word
conduct ：v. 实施, 执行, 表现, 引导 n. 举止, 管理方法, 经营方式, 实施办法
qualitative results：定性结果
intuitive view：直观的看法
validity ：有效性 真实性, 效力, 有效

4.2.1 Comparisons with State-of-the-art Methods 与最先进方法的比较

In Table 1, we compare our method with former state-of-the-art methods which mostly report results with multiple random runs. Our proposed DCNet achieves state-of-the-art results on almost all the splits with different shots and outperforms previous methods by a large margin. Specifically, in extremely low-shot settings (i.e. 1-shot), our method outperforms others by about 10% in split 1 and 3, providing a convincing proof that our DCNet is able to capture local detailed information to overcome the variations brought by the randomly sampled training shots.

在表1中，我们将我们的方法与以前最先进的方法进行了比较，这些方法大多报告了多次随机运行的结果。我们提出的DCNet在几乎所有不同镜头的分片上都取得了最先进的结果，并以很大的优势胜过了以前的方法。具体来说，在极低的镜头设置（即1个镜头）中，我们的方法在分割1和分割3中比其他方法高出10%，提供了一个令人信服的证据，即我们的DCNet能够捕捉局部的详细信息，以克服随机抽样训练样本带来的变化。

Word
convincing ：形容词: 使人信服
by a large margin：大幅度
proof ：n. 证明, 证据, 检验, 证实 adj. 能抵御, 能防范, 可防护, 防…的 v. 给（织物等）做防护处理, 使防水（或防火等）, 印…的校样

Table 1. Few-shot object detection performance on VOC 2007 test set of PASCAL VOC dataset. We report the mAP with IoU threshold 0.5 (AP50) under three different splits for five novel classes. * denotes the results averaged over multiple random runs.
表 1. PASCAL VOC 数据集的 VOC 2007 测试集上的小样本目标检测性能。我们报告了在五个新类别的三个不同分割下 IoU 阈值为 0.5 (AP50) 的 mAP。 * 表示多次随机运行的平均结果。

4.2.2 Ablation Study 消融研究

We present results of comprehensive ablation studies to analyze the effectiveness of various components of the proposed DCNet. All ablation studies are conducted on the PASCAL VOC 2007 test set with the first novel splits. All results are averaged over 10 random runs.

我们展示了综合消融研究的结果，以分析所提出的 DCNet 的各个组件的有效性。所有的消融研究都是在 PASCAL VOC 2007 测试集上进行的，并带有第一个新的拆分。所有结果都是 10 次随机运行的平均值。

Impact of dense relation distillation module. We conduct experiments to validate the superiority of the proposed dense relation distillation (DRD) module. Specifically, we implement the baseline method for meta-learning based few-shot detection Meta R-CNN with class-specific prediction for the final box classification and regression. While the DRD module requires no extra class-specific processing. As shown in line 1 and 2 of Table 2, DCNet w/o CFA equals to Faster R-CNN equipped with DRD module, our proposed DRD module achieves consistent improvement on all novel splits with all shots number, which effectively demonstrates the supremacy of the relation distillation mechanism over the baseline method. Moreover, the improvement over baseline is significant when the shot number is low, which proves that the DRD module successfully exploits useful information from limited support data.

密集关系蒸馏模块的影响。 我们进行实验以验证所提出的密集关系蒸馏（DRD）模块的优越性。具体来说，我们实现了基于元学习的小样本检测 Meta R-CNN 的基线方法，并针对最终框分类和回归进行了特定类别的预测。而 DRD 模块不需要额外的特定于类的处理。如表 2 的第 1 行和第 2 行所示，DCNet w/o CFA 等于配备 DRD 模块的 Faster R-CNN，我们提出的 DRD 模块在所有镜头数的所有新分裂上都实现了一致的改进，这有效地证明了基线方法上的关系蒸馏机制。此外，当样本数较低时，对基线的改进是显着的，这证明 DRD 模块成功地利用了有限支持数据中的有用信息。

Word
validate ：证实
the superiority of：的优越性
Specifically：adv. 具体来说, 明确地, 具体地, 特意
extra：额外的 额外, 剩余的, 补充
the supremacy of：至高无上

Table 2. Ablation study to evaluate the effectiveness of different components in our proposed method. The mAP with IoU threhold 0.5 (AP50) is reported. * denotes CFA module with attention aggregation fashion. † denotes our implementation.
表 2. 评估我们提出的方法中不同组件有效性的消融研究。报告了 IoU 阈值为 0.5 (AP50) 的 mAP。 * 表示具有注意力聚合方式的 CFA 模块。 † 表示我们的实现。

Impact of context-aware feature aggregation module. We carry out experiments to evaluate the validity of the proposed context-aware feature aggregation (CFA) module. Specifically, RoI features generated from parallel branches are aggregated with a simple summation. From line 1 and 3 of the table, with the introduction of CFA module, Meta R-CNN achieves notable gains over the baseline. Since CFA module targets at preserving detailed information in a scale-aware manner, different levels of detailed features can be retrieved to assist the prediction process.

上下文感知特征聚合模块的影响。 我们进行了实验来评估所提出的上下文感知特征聚合（CFA）模块的有效性。具体来说，从并行分支生成的 RoI 特征通过简单的求和进行聚合。从表的第 1 行和第 3 行来看，随着 CFA 模块的引入，Meta R-CNN 在基线上取得了显着的进步。由于 CFA 模块的目标是以尺度感知的方式保存详细信息，因此可以检索不同级别的详细特征以帮助预测过程。

Word
carry out :. 实行, 开展, 完成
notable :n. 名人, 重要人物 adj. 值得注意的, 显著的, 重要的
assist :帮助 协助, 辅助, 辅佐

Impact of different RoI pooling resolutions. To further evaluate the impact of different RoI pooling resolutions, we perform explicit experiments to show the detailed performance. As shown in Table 3, solely adopting larger pooling resolution could yield better performance. However, only when aggregating features generated with all three resolutions, the best performance could be obtained.

不同 RoI 池化分辨率的影响。 为了进一步评估不同 RoI 池化分辨率的影响，我们进行了明确的实验以显示详细的性能。如表 3 所示，单独采用更大的池化分辨率可以获得更好的性能。但是，只有聚合所有三种分辨率生成的特征时，才能获得最佳性能。

Word
explicit :adj. 清楚明白的, 易于理解的, 明确的, 直言的
solely :adv. 只, 仅, 唯, 单独地
yield :n. 产量, 产出, 利润 v. 屈服, 让步, 放弃, 提供

Table 3. The impact of different RoI pooling resolutions. The experiments are conducted on VOC 2007 test set of PASCAL VOC dataset with novel split1 and AP50 on 10-shot task averaged from 10 random runs is reported.
表 3. 不同 RoI 池化分辨率的影响。实验是在 PASCAL VOC 数据集的 VOC 2007 测试集上进行的，报告了新的 split1 和 AP50 对 10 次随机运行的平均值的 10-shot 任务。

Impact of attentive aggregation fashion for CFA module. Based on the plain CFA module, we further propose an attention-based aggregation mechanism to adaptively fuse different RoI features. As presented in line 3 and line 4 of Table 2, the attention aggregation mechanism can further boost the performance of the model, which promotes the plain CFA module with a more comprehensive feature representation, effectively balancing the contributions of each extracted features. Finally, with the combination of DRD module and CFA module, we present DCNet, which achieves the best performance according to Table 2.

注意力聚合方式对 CFA 模块的影响。 基于普通的 CFA 模块，我们进一步提出了一种基于注意力的聚合机制，以自适应地融合不同的 RoI 特征。如表 2 的第 3 行和第 4 行所示，注意力聚合机制可以进一步提升模型的性能，从而促进具有更全面特征表示的普通 CFA 模块，有效平衡每个提取特征的贡献。最后，结合 DRD 模块和 CFA 模块，我们提出了 DCNet，它根据表 2 实现了最佳性能。

4.2.3 Qualitative Results 定性结果

To further comprehend the effect of dense relation distillation (DRD) module, we visualize features before and after DRD module. As shown in Fig. 5 (a), after relation distillation, query features can be activated to facilitate the subsequent detection procedure. Moreover, different from former meta-learning based methods which performs prediction in a class-wise manner, our proposed DRD module can model relations between query and support features in all classes at the same time as shown in the second line of Fig. 5 (a). The DRD module enables the model to focus more on the query objects under the guidance of support information. Additionally, we also visualize the effect of CFA module presented in Fig. 5 (b). With a relatively large or small query object as input, DCNet w/o CFA suffers from false classification or missing detection , while the introduction of CFA module could effectively resolve this issue.

为了进一步理解密集关系蒸馏 (DRD) 模块的效果，我们将 DRD 模块前后的特征可视化。如图5（a）所示，在关系蒸馏之后，可以激活查询特征以方便后续的检测过程。此外，与以前以类方式执行预测的基于元学习的方法不同，我们提出的 DRD 模块可以同时建模所有类中查询和支持特征之间的关系，如图 5 的第二行所示（一种）。 DRD 模块使模型在支持信息的指导下更加关注查询对象。此外，我们还可视化了图 5（b）中呈现的 CFA 模块的效果。以相对较大或较小的查询对象作为输入，DCNet w/o CFA 存在误分类或漏检的问题，而引入 CFA 模块可以有效解决这个问题。

Word
facilitate :v. 促进, 促使, 使便利
subsequent :adj. 随后的, 后来的, 之后的, 接后的

4.3. Experiments on MS COCO MS COCO 上的实验

We evaluate 10/30-shot setups on MS COCO benchmark and report the averaged performance with the standard COCO metrics over 10 runs with random shots. The results on novel classes can be seen in Table 4. Despite the challenging nature of COCO dataset with large number of categories, our proposed DCNet achieves state-of-the-art performance on most of the metrics.

我们在 MS COCO 基准测试中评估 10/30-shot 设置，并报告使用标准 COCO 指标在 10 次随机测试中运行的平均性能。新类别的结果可以在表 4 中看到。 尽管 COCO 数据集具有大量类别的挑战性，但我们提出的 DCNet 在大多数指标上都实现了最先进的性能。

Table 4. Few-shot object detection performance on COCO minival of MS COCO dataset. We report the mean Averaged Precision and mean Averaged Recall on the 20 novel classes of COCO. * denotes the results averaged over multiple random runs.
表 4. MS COCO 数据集的 COCO minival 上的小样本目标检测性能。我们报告了 20 个新型 COCO 的平均平均精度和平均平均召回率。 * 表示多次随机运行的平均结果。

Figure 5. (a). Visualizations of features before and after dense relation distillation module. (b). Visualizations of effect of context-aware feature aggregation module.
图 5. (a)。密集关系蒸馏模块前后特征的可视化。 (b)。上下文感知特征聚合模块效果的可视化。

5. Conclusions 结论

In this paper, we have presented the Dense Relation Distillation Network with Context-aware Aggregation (DCNet) to tackle few-shot object detection problem. Dense relation distillation module adopts dense matching strategy between query and support features to fully exploit support information. Furthermore, context-aware feature aggregation module adaptively harnesses features from different scales to produce a more comprehensive feature representation. The ablation experiments demonstrate the effectiveness of each component of DCNet. Our proposed DCNet achieves state-of-the-art results on two benchmark datasets, i.e. PASCAL VOC and MS COCO.

在本文中，我们提出了具有上下文感知聚合（DCNet）的密集关系蒸馏网络来解决少镜头对象检测问题。密集关系蒸馏模块采用查询和支持特征之间的密集匹配策略来充分利用支持信息。此外，上下文感知特征聚合模块自适应地利用不同尺度的特征来产生更全面的特征表示。消融实验证明了 DCNet 的每个组件的有效性。我们提出的 DCNet 在两个基准数据集上取得了最先进的结果，即 PASCAL VOC 和 MS COCO。