Abstract:With the development of convolutional neural networks, significant breakthroughs have been made in deep learning-based target detection algorithms. However, existing target detection algorithms based on convolutional neural networks need to downsample the whole image to extract deep semantic information from the image, which can lead to the loss of spatial information for small targets, impressive results on large/medium-sized targets, the results are not satisfactory for small target detection. To solve the problem of small target detection and increase the detection precision of small targets, we propose an end-to-end generative adversarial network GAN-STD for small target detection in this paper. GAN-STD makes full use of the structural correlation between targets at different scales through generative adversarial networks to enhance the similar representation of small targets in the shallow feature map and large targets in the deep feature map in the feature extraction process, and reduce the difference between the representation of small targets and large targets, thus making small targets as easy to detect as large targets. In addition, for the detector to perform better localization and classification of small targets, we back-propagate the detector losses to the generator and discriminator for end-to-end training. We merge GAN-STD onto two widely used one-stage target detectors (SSD and YOLOv4) to validate the effectiveness of our proposed GAN-STD. Extensive experiments on the widely used PASCAL VOC, MS COCO, and TT100K datasets show that the proposed GAN-STD algorithm achieves excellent results for detecting small targets.

It GAN Do Better: GAN-Based Detection of Objects on Images With Varying Quality

Online Adversarial Knowledge Distillation for Image Synthesis of Bridge Defect

Low Illumination Enhancement for Object Detection in Self-Driving

No One Can Escape: A General Approach to Detect Tampered and Generated Image

An enhanced algorithm for object detection based on generative adversarial structure

RodNet: An Advanced Multi-Domain Object Detection Approach using Feature Transformation with Generative Adversarial Networks

Chosen methods of improving small object recognition with weak recognizable features

Improved Object Detection using Data Enhancement method based on Generative Adversarial Nets

Robust Data Augmentation Generative Adversarial Network for Object Detection

Single-Shot Bidirectional Pyramid Networks for High-Quality Object Detection.

Small-Object Detection in Remote Sensing Images with End-to-End Edge-Enhanced GAN and Object Detector Network

Improve Object Detection by Data Enhancement based on Generative Adversarial Nets

Towards Robust GAN-generated Image Detection: a Multi-view Completion Representation

A Generative Adversarial Network Based Deep Learning Method for Low-Quality Defect Image Reconstruction and Recognition

Domain Adaptive Object Detection with Generative Adversarial Network

Perceptual Generative Adversarial Networks for Small Object Detection

A Mixed-Supervision Multilevel GAN Framework for Image Quality Enhancement

T-GD: Transferable GAN-generated Images Detection Framework

Improving GAN Training via Feature Space Shrinkage

GAN-STD: small target detection based on generative adversarial network

Defense-GAN: Protecting Classifiers Against Adversarial Attacks Using Generative Models