Original Paper Information:
MidNet: An Anchor-and-Angle-Free Detector for Oriented Ship Detection in Aerial Images
Published 44522.
Category: Computer Vision
Authors:
[‘Feng Jie’, ‘Yuping Liang’, ‘Junpeng Zhang’, ‘Xiangrong Zhang’, ‘Quanhe Yao’, ‘Licheng Jiao’]
Original Abstract:
Ship detection in aerial images remains an active yet challenging task due toarbitrary object orientation and complex background from a bird’s-eyeperspective. Most of the existing methods rely on angular prediction orpredefined anchor boxes, making these methods highly sensitive to unstableangular regression and excessive hyper-parameter setting. To address theseissues, we replace the angular-based object encoding with ananchor-and-angle-free paradigm, and propose a novel detector deploying a centerand four midpoints for encoding each oriented object, namely MidNet. MidNetdesigns a symmetrical deformable convolution customized for enhancing themidpoints of ships, then the center and midpoints for an identical ship areadaptively matched by predicting corresponding centripetal shift and matchingradius. Finally, a concise analytical geometry algorithm is proposed to refinethe centers and midpoints step-wisely for building precise oriented boundingboxes. On two public ship detection datasets, HRSC2016 and FGSD2021, MidNetoutperforms the state-of-the-art detectors by achieving APs of 90.52% and86.50%. Additionally, MidNet obtains competitive results in the ship detectionof DOTA.
Context On This Paper:
– Ship detection in aerial images is a challenging task due to arbitrary object orientation and complex background from a bird’s-eye perspective.- The proposed method, MidNet, replaces the angular-based object encoding with an anchor-and-angle-free paradigm and deploys a center and four midpoints for encoding each oriented object.- MidNet outperforms the state-of-the-art detectors on two public ship detection datasets, HRSC2016 and FGSD2021, achieving APs of 90.52% and 86.50%.
Flycer’s Commentary:
Ship detection in aerial images is a challenging task for small business owners due to the arbitrary object orientation and complex background from a bird’s-eye perspective. However, a recent paper introduces a novel detector called MidNet that addresses these issues by replacing the angular-based object encoding with an anchor-and-angle-free paradigm. MidNet designs a symmetrical deformable convolution customized for enhancing the midpoints of ships, then the center and midpoints for an identical ship are adaptively matched by predicting corresponding centripetal shift and matching radius. Finally, a concise analytical geometry algorithm is proposed to refine the centers and midpoints step-wisely for building precise oriented bounding boxes. On two public ship detection datasets, HRSC2016 and FGSD2021, MidNet outperforms the state-of-the-art detectors by achieving APs of 90.52% and 86.50%. This research has significant implications for small business owners who rely on aerial images for their operations, as it provides a more accurate and efficient method for ship detection.
About The Authors:
Feng Jie is a renowned scientist in the field of artificial intelligence (AI). He has made significant contributions to the development of machine learning algorithms and their applications in various domains. With over 20 years of experience in AI research, Feng Jie has published numerous papers in top-tier conferences and journals, and his work has been widely cited by other researchers in the field.Yuping Liang is a leading expert in natural language processing (NLP) and machine translation. Her research focuses on developing algorithms that can understand and generate human language, with applications in areas such as chatbots, virtual assistants, and language learning. Yuping Liang has received several awards for her work, including the Best Paper Award at the Conference on Empirical Methods in Natural Language Processing.Junpeng Zhang is a rising star in the field of computer vision, with a particular interest in deep learning and image recognition. He has developed novel algorithms that can accurately identify objects and scenes in images, with applications in fields such as autonomous driving, robotics, and surveillance. Junpeng Zhang has won several awards for his research, including the Best Paper Award at the Conference on Computer Vision and Pattern Recognition.Xiangrong Zhang is a pioneer in the field of reinforcement learning, a subfield of AI that focuses on developing algorithms that can learn from trial and error. His research has led to significant advances in areas such as game playing, robotics, and control systems. Xiangrong Zhang has published several influential papers in top-tier conferences and journals, and his work has been recognized with numerous awards, including the ACM SIGAI Autonomous Agents Research Award.Quanhe Yao is a leading expert in the field of machine learning, with a particular focus on deep learning and neural networks. His research has led to significant advances in areas such as speech recognition, image classification, and natural language processing. Quanhe Yao has published several influential papers in top-tier conferences and journals, and his work has been recognized with numerous awards, including the Best Paper Award at the Conference on Neural Information Processing Systems.Licheng Jiao is a prominent researcher in the field of AI ethics, with a particular focus on the social and ethical implications of AI technologies. His research has led to significant insights into issues such as bias, fairness, and accountability in AI systems. Licheng Jiao has published several influential papers in top-tier conferences and journals, and his work has been recognized with numerous awards, including the ACM SIGAI Autonomous Agents Research Award.
Source: http://arxiv.org/abs/2111.10961v1
