YOLOX检测器的锚框:采用anchor-free的方式,以及使用例如网络头解耦(decoupled head),与以标签分配策略为主导(leading label assignment strategy SimOTA)的技术生成网络模型。
YOLOX使用的相关技术论文
1. anchor-free detectors:
(1) Hei Law and Jia Deng. Cornernet: Detecting objects as paired keypoints. In ECCV, 2018.
(2) Zhi Tian, Chunhua Shen, Hao Chen, and Tong He. Focs: Fully convolutional one-stage object detection. In ICCV, 2019.
(3) Xingyi Zhou, Dequan Wang, and Philipp Krahenbuhl. Objects as points. arXiv preprint arXiv:1904.07850, 2019.
最开始的anchor-free:DenseBox、YOLO。两则比较相似,都是单阶段(one-stege)的目标检测。区别在于DenseBox为密集预测,基于每个像素做预测,YOLO则以划分的网格做预测。
2. advanced label assignment strategies:
(1) Zheng Ge, Songtao Liu, Zeming Li, Osamu Yoshie, and Jian Sun. Ota: Optimal transport assignment for object detection. In CVPR, 2021.
(2) Kang Kim and Hee Seok Lee. Probabilistic anchor assignment with iou prediction for object detection. In ECCV, 2020.
(3) Yuchen Ma, Songtao Liu, Zeming Li, and Jian Sun. Iqdet: Instance-wise quality distribution sampling for object detection. In CVPR, 2021.
(4) Shifeng Zhang, Cheng Chi, Yongqiang Yao, Zhen Lei, and Stan Z Li. Bridging the gap between anchor-based and anchor-free detection via adaptive training sample selection. In CVPR, 2020.
(5) Xiaosong Zhang, Fang Wan, Chang Liu, Rongrong Ji, and Qixiang Ye. Freeanchor: Learning to match anchors for visual object detection. In NeurIPS, 2019.
(6) Benjin Zhu, Jianfeng Wang, Zhengkai Jiang, Fuhang Zong, Songtao Liu, Zeming Li, and Jian Sun. Autoassign: Differentiable label assignment for dense object detection. arXiv preprint arXiv:2007.03496, 2020.
3. end-to-end(NMS-free) detector:
(1) Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-toend object detection with transformers. In ECCV, 2020.
(2) Jianfeng Wang, Lin Song, Zeming Li, Hongbin Sun, Jian Sun, and Nanning Zheng. End-to-end object detection with fully convolutional network. In CVPR, 2020.
(3) Qiang Zhou, Chaohui Yu, Chunhua Shen, Zhibin Wang, and Hao Li. Object detection made simpler by eliminating heuristic nms. arXiv preprint arXiv:2101.11782, 2021.
网络头解耦(Decoupled head)
YOLOX的实验表明解耦的检测头会带来如下影响:
(1) 能够提升训练的收敛速度
(2) 解耦是end2end的要点。
通过一幅图认知YOLOv3 head与Decoupled head的区别,也展示了Decoupled Head。