本文提出了一种新颖的摆脱锚框的椭圆模板检测器EllipseNet，其能够直接预测椭圆框的圆心，长短轴长以及旋转角度。由于椭圆模板检测网络能够自适应的调整旋转角度故能够更加精确的检测出物体的轮廓。另外，为了提高模型的检测精度，设计了一种高效的椭圆交并比损失函数。实验结果表明，由于没有引入复杂的锚框，该模型结构简单，且取得了和基于矩形框目标检测算法相近的性能。

This paper propose a novel anchor-free object detector, namely EllipseNet. Instead of using the rectangular box, EllipseNet detects an elliptical box by predicting the center, axis and tilt angle of the ellipse. As the ellipse model can capture the object rotation and match the target adaptively by adjusting tilt angle, EllipseNet has more accurate detection boundary. In addition, we introduce an Elliptical IOU Loss to improve the detection performance of the model. The experimental results show that the model has simple structure and achieves similar performance to the object detection methods based on rectangular boxes because there is no complex anchor boxes introduced.