DetNet: A Backbone network for Object Detection
Zeming Li1 Chao Peng2 Gang Yu2 Xiangyu Zhang2 Yangdong Deng1 Jian Sun2 1School of Software Tsinghua University } 2 Megvii Inc. (Face++) {pengchao yugang zhangxiangyu
Abstract
Recent CNN based object detectors, no matter one-stage methods like YOLO [1, 2], SSD [3], and RetinaNet [4] or two-stage detectors like Faster R-CNN [5], R-FCN [6] and FPN [7] are usually trying to directly finetune from ImageNet pre-trained models designed for image classification. There has been little work discussing on the backbone feature extractor specifically designed for the object detection. More importantly, there are several differences between the tasks of image classification and object detection. (i) Recent object detectors like FPN and RetinaNet usually involve extra stages against the task of image classification to handle the objects with various scales. (ii) Object detection not only needs to recognize the category of the object instances but also spatially locate the position. Large downsampling factor brings large valid receptive field, which is good for image classification but compromises the object location ability. Due to the gap between the image classification and object detection, we propose DetNet in this paper, which is a novel backbone network specifically designed for object detection. Moreover, DetNet includes the extra stages against traditional bac
中文速览
目标检测任务长期借用为图像分类设计的骨干网络（如ResNet），但分类和检测之间存在两大根本矛盾：检测器需要比分类网络更多的特征提取阶段来应对多尺度目标，而分类网络为扩大感受野采用的大步长下采样又会损害空间分辨率，导致大目标边界模糊、小目标信息丢失。针对这些问题，研究者提出了专为目标检测设计的骨干网络DetNet：它在ResNet基础上直接增加检测所需的额外阶段（如P6），并通过引入空洞卷积（dilated convolution）瓶颈结构，在不增加下采样的前提下保持高空间分辨率，同时维持足够大的感受野。在MS COCO基准上，DetNet59仅用4.8G浮点运算量就在目标检测和实例分割两项任务上达到了当时的最优性能。这项工作的意义在于首次系统指出"分类骨干网络天然不适合检测任务"这一问题，并给出了一个轻量高效的专用解决方案，为后续检测骨干网络的定制化设计提供了重要参考。
原文 arXiv:1804.06215;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1804.06215v2