table of Contents
Abstract
Currently, many good results of object detection algorithms are enumerated a number of possible locations and classify them one by one, these methods are not only a waste of resources, but also post-processing (generally refers NMS), it is very inefficient. In this paper, we use a different algorithm - will be tested to predict a target point (the border of the center of the target). Our prediction algorithm using the key point to find the center point, and return Further features of the object, such as size, 3D position, orientation, and even posture. Compared to those based detector Bounding box, our CenterNet end to end, simpler, faster and more accurate detector.
Instruction
Analysis of Loss of formula CenterNet
\[ L_{det} = L_k+\lambda_{size}L_{size}+\lambda_{off}L_{off} \]
The first part: \ (L_K \)
