TY - GEN
T1 - Guided refine-head for object detection
AU - Zeng, Lingyun
AU - Song, You
AU - Wang, Wenhai
N1 - Publisher Copyright:
© 2020, Springer Nature Switzerland AG.
PY - 2020
Y1 - 2020
N2 - In recent years, multi-stage detectors improve the accuracy of object detection to a new level. However, due to multiple stages, these methods typically fall short in the inference speed. To alleviate this problem, we propose a novel object detector—Guided Refine-Head, which is made up of a newly proposed detection network called Refine-Head and a knowledge-distillation-like loss function. Refine-Head is a two-stage detector, and thus Refine-Head has faster inference speed than multi-stage detectors. Nonetheless, Refine-Head is able to predict bounding boxes for incremental IoU thresholds like a multi-stage detector. In addition, we use knowledge-distillation-like loss function to guide the training process of Refine-Head. Therefore, besides fast inference speed, the proposed Guided Refine-Head also has competitive accuracy. Abundant ablation studies and comparative experiments on MS-COCO 2017 validate the superiority of the proposed Guided Refine-Head. It is worth noting that Guided Refine-Head achieves the AP of 38.0% at 10.4 FPS, surpassing Faster R-CNN by 1.8% at the similar speed.
AB - In recent years, multi-stage detectors improve the accuracy of object detection to a new level. However, due to multiple stages, these methods typically fall short in the inference speed. To alleviate this problem, we propose a novel object detector—Guided Refine-Head, which is made up of a newly proposed detection network called Refine-Head and a knowledge-distillation-like loss function. Refine-Head is a two-stage detector, and thus Refine-Head has faster inference speed than multi-stage detectors. Nonetheless, Refine-Head is able to predict bounding boxes for incremental IoU thresholds like a multi-stage detector. In addition, we use knowledge-distillation-like loss function to guide the training process of Refine-Head. Therefore, besides fast inference speed, the proposed Guided Refine-Head also has competitive accuracy. Abundant ablation studies and comparative experiments on MS-COCO 2017 validate the superiority of the proposed Guided Refine-Head. It is worth noting that Guided Refine-Head achieves the AP of 38.0% at 10.4 FPS, surpassing Faster R-CNN by 1.8% at the similar speed.
KW - Knowledge distillation
KW - Object detection
KW - Two-stage detector
UR - https://www.scopus.com/pages/publications/85078528428
U2 - 10.1007/978-3-030-37731-1_17
DO - 10.1007/978-3-030-37731-1_17
M3 - 会议稿件
AN - SCOPUS:85078528428
SN - 9783030377304
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 203
EP - 214
BT - MultiMedia Modeling - 26th International Conference, MMM 2020, Proceedings
A2 - Cheng, Wen-Huang
A2 - Kim, Junmo
A2 - Choi, Jung-Woo
A2 - Chu, Wei-Ta
A2 - Cui, Peng
A2 - Hu, Min-Chun
A2 - De Neve, Wesley
PB - Springer
T2 - 26th International Conference on MultiMedia Modeling, MMM 2020
Y2 - 5 January 2020 through 8 January 2020
ER -