Research Area:  Machine Learning

Abstract:

With the rapid development of deep learning, the accuracy of object detection has greatly improved. But like the dark clouds in the clear sky, small objects detection is still a problem in computer vision. Due to low resolution, few pixels, limited information and rough outline, the features are difficult to extract, and training data is difficult to label. Around the problems and challenges in small object detection, we made an analysis and overview of the research progress and status of deep learning-based algorithms from one-stage and two-stage object detection methods. Then some techniques for small object detection are analysed from the aspects of feature enhancement, multi-scale prediction and detection network model. Finally, aiming at the problems and the development trend of machine vision, we make predictions and prospects for the application and development of small object detection in the future.

Keywords:  
Deep Learning
Small Object Detection
multi-scale prediction
Deep Learning
Machine Learning

Author(s) Name:  Yingjie Liang , Yueying Han , Feng Jiang

Publisher name:  ACM

DOI:  10.1145/3532213.3532278

Volume Information:  

Paper Link:   https://dl.acm.org/doi/abs/10.1145/3532213.3532278