3D object detection is an essential step towards holistic scene understanding. Currently, the existing 3D object detection methods focus on certain object’s areas once and predict the object’s locations. The way does not conform to the habit of hu...
3D object detection is an essential step towards holistic scene understanding. Currently, the existing 3D object detection methods focus on certain object’s areas once and predict the object’s locations. The way does not conform to the habit of human observing targets. Hence, this work proposes a fast and accurate object detector called 3D SaccadeNet, which regards one 3D object as nine keypoints. In the training process, the corner loss, center loss, and classification loss are computed. However, the center is only used to predict a 3D object. Performed experiments on the KITTI dataset show that the proposed method is highly efficient and effective, and the 3D object detection reaches (91:18%; 82:80%; 79:90%).