Abstract: The deep CNN-based object detector has achieved a remarkable grade in recent year, which far exceeds the traditional counterparts. However, the near researches show that the feature mismatch problem has become a bottleneck to further improve performance. Consider a clear summary on the problem has not yet been collected, this paper firstly takes a deep analysis. And points out that that feature mismatch is essentially caused by the different optimization objectives of classification task and regression task in object detection. Meanwhile, it also analyses the impact of the problem on dense prediction strategy and believes that it leads to response mismatch of the two tasks. And then, the exists solutions of feature mismatch and response mismatch are introduced in detail respectively. At last, a conclusion is drew that response mismatch should be considered simultaneously when solving the feature mismatch. And some ideas are given to solve the problems all in one.