Scale-Aware Trident Networks for Object Detection
Yanghao Li* Yuntao Chen
1,3
* Naiyan Wang
2
Zhaoxiang Zhang
1,3,4
1
University of Chinese Academy of Sciences
2
TuSimple
3
Center for Research on Intelligent Perception and Computing, CASIA
4
Center for Excellence in Brain Science and Intelligence Technology, CAS
lyttonhao@gmail.com chenyuntao2016@ia.ac.cn zhaoxiang.zhang@ia.ac.cn winsty@gmail.com
Abstract
Scale variation is one of the key challenges in object de-
tection. In this work, we first present a controlled experi-
ment to investigate the effect of receptive fields on the detec-
tion of different scale objects. Based on the findings from the
exploration experiments, we propose a novel Trident Net-
work (TridentNet) aiming to generate scale-specific feature
maps with a uniform representational power. We construct
a parallel multi-branch architecture in which each branch
shares the same transformation parameters but with differ-
ent receptive fields. Then, we propose a scale-aware train-
ing scheme to specialize each branch by sampling object
instances of proper scales for training. As a bonus, a fast
approximation version of TridentNet could achieve signifi-
cant improvements without any additional parameters and
computational cost. On the COCO dataset, our TridentNet
with ResNet-101 backbone achieves state-of-the-art single-
model results by obtaining an mAP of 48.4. Code will be
made publicly available.
1. Introduction
In recent years, deep convolutional neural networks
(CNNs) [17, 37, 30] have achieved great success in ob-
ject detection. Typically, these CNN-based methods can be
roughly divided into two types: one stage methods such
as YOLO [34] or SSD [30] which directly utilizes feed-
forward CNN to predict the bounding boxes of interest,
while two stage methods such as Faster R-CNN [37] or
R-FCN [10] first generate proposals, and then exploit the
extracted region features from CNN for further refinement.
However, a central issue in both methods lies in handling
scale variation. It is very common that the scale of object
instances varies in a wide range, which impedes the detec-
tors, especially for very small or very large objects.
* Equal Contribution
To remedy the scale variation issue, an intuitive way is
to leverage multi-scale image pyramids [1], which is pop-
ular in both hand-crafted feature based methods [12, 31]
and current deep CNN based methods (Figure 1(a)). Strong
evidence [22, 29] shows that current standard deep detec-
tors [37, 10] could benefit from multi-scale training and
testing. To avoid training objects with extreme scales
(small/large objects in smaller/larger scales), SNIP [40, 41]
proposes a scale normalization method that selectively
trains the objects of appropriate sizes in each image scale.
Nevertheless, the increase of inference time makes the im-
age pyramid methods infeasible for practical applications.
The other line of efforts aims to employ in-network fea-
ture pyramids to approximate image pyramids with less
computation cost. The idea is first demonstrated in [13],
where a fast feature pyramid is constructed for object de-
tection by interpolating some feature channels from nearby
scale levels. In the deep learning era, the approximation
is even easier. SSD [30] utilizes multi-scale feature maps
from different layers and detects objects of different scales
at each feature layer. To compensate the absence of seman-
tics in low-level features, FPN [26] (Figure 1(b)) further
augments a top-down pathway and lateral connections to
incorporate strong semantic information in high-level fea-
tures. However, the representational power for objects of
different scales still differ, since their features are extracted
on different layers in FPN. This makes feature pyramids an
unsatisfactory alternative for image pyramids.
Both image pyramid and feature pyramid methods share
the same motivation that models should have different re-
ceptive fields for objects of different scales. Despite of
the inefficiency, image pyramids fully utilize the representa-
tional power of the model to transform objects of all scales
equally. In contrast, feature pyramids generate multi-level
features thus sacrificing the feature consistency across dif-
ferent scales. The goal of this work is to get the best of two
worlds by creating features with a uniform representational
power for all scales efficiently.
In this paper, instead of feeding in multi-scale inputs
1
arXiv:1901.01892v1 [cs.CV] 7 Jan 2019
下载后可阅读完整内容，剩余9页未读，立即下载
nihate
- 粉丝: 1728
- 资源: 24
我的内容管理
展开
最新资源
- 李兴华Java基础教程：从入门到精通
- U盘与硬盘启动安装教程：从菜鸟到专家
- C++面试宝典：动态内存管理与继承解析
- C++ STL源码深度解析：专家级剖析与关键技术
- C/C++调用DOS命令实战指南
- 神经网络补偿的多传感器航迹融合技术
- GIS中的大地坐标系与椭球体解析
- 海思Hi3515 H.264编解码处理器用户手册
- Oracle基础练习题与解答
- 谷歌地球3D建筑筛选新流程详解
- CFO与CIO携手：数据管理与企业增值的战略
- Eclipse IDE基础教程：从入门到精通
- Shell脚本专家宝典：全面学习与资源指南
- Tomcat安装指南：附带JDK配置步骤
- NA3003A电子水准仪数据格式解析与转换研究
- 自动化专业英语词汇精华：必备术语集锦
资源上传下载、课程学习等过程中有任何疑问或建议，欢迎提出宝贵意见哦~我们会及时处理！
点击此处反馈
