Reading Note: DetNet: A Backbone network for Object Detection

TITLE: DetNet: A Backbone network for Object Detection

AUTHOR: Xuepeng Shi, Shiguang Shan, Meina Kan, Shuzhe Wu, Xilin Chen

ASSOCIATION: Tsinghua University, Face++

FROM: arXiv:1804.06215

CONTRIBUTION

1. The inherent drawbacks of traditional ImageNet pre-trained model for fine-tunning recent object detectors is analyzed.
2. A novel backbone, called DetNet, is proposed, which is specifically designed for object detection task by maintaining the spatial resolution and enlarging the receptive field.

METHOD

Motivation

There are two problems using the classification backbone for object detection tasks. (i) Recent detectors, e.g., FPN, involve extra stages compared with the backbone network for ImageNet classification in order to detect objects with various sizes. (ii) Traditional backbone produces higher receptive field based on large downsampling factor, which is beneficial to the visual classification. However, the spatial resolution is compromised which will fail to accurately localize the large objects and recognize the small objects.

To sumarize, there are 3 main problems to use current pre-trained models, including

1. The number of network stages is different. It means that extra layers for object detection compared to classification has not been pretrained.
2. Weak visibility of large objects. It is because The feature map with strong semantic information has large strides respect to input image, which is harmful for the object localization.
3. Invisibility of small objects. The information from the small objects will be easily weaken as the spatial resolution of the feature maps is decreased and the large context information is integrated.

To address these problems, DetNet has following characteristics. (i) The number of stages is directly designed for Object Detection. (ii) Even though more stages are involved, high spatial resolution of the feature maps is mainted, while keeping large receptive field using dilated convolution.

DetNet Design

The main architecture of DetNet is designed based on ResNet-50. The first 4 stages are kept same with ResNet-50. The main differences are illustrated as follows:

1. The extra stages are merged into the backbone which will be later utilized for object detection as in FPN. Meanwhile, the spatial resolution is fixed as 16x downsampling even after stage 4.
2. Since the spatial size is fixed after stage 4, in order to introduce a new stage, a dilated bottleneck with $1 \times 1$ convolution projection is utilized in the begining of the each stage. The dilation convolution efficiently enlarge the receptive field.
3. Since dilated convolution is still time consuming, stage 5 and stage 6 keep the same channels as stage 4 (256 input channels for bottleneck block). This is different from traditional backbone design, which will double channels in a later stage.

The following figure shows the dialted bottleneck with $1 \times 1$ conv projection and the architecture of DetNet.

PERFORMANCE