The paper introduces ActiveAnno3D, an active learning framework designed to reduce the annotation costs in multi-modal 3D object detection. The framework aims to select the most informative data samples for labeling, thereby improving detection performance with fewer annotations. The authors explore various continuous training methods and integrate the most efficient one, demonstrating that their approach can achieve comparable performance to full data annotation with significantly less labeling effort. Extensive experiments are conducted on the nuScenes and TUM Traffic Intersection datasets using BEVFusion and PV-RCNN models. The results show that ActiveAnno3D can achieve high mAP scores with only 50% of the training data, making it a promising solution for efficient and accurate 3D object detection in autonomous driving applications. The framework is also integrated into the proAnno labeling tool to enable AI-assisted data selection and minimize manual labeling costs.The paper introduces ActiveAnno3D, an active learning framework designed to reduce the annotation costs in multi-modal 3D object detection. The framework aims to select the most informative data samples for labeling, thereby improving detection performance with fewer annotations. The authors explore various continuous training methods and integrate the most efficient one, demonstrating that their approach can achieve comparable performance to full data annotation with significantly less labeling effort. Extensive experiments are conducted on the nuScenes and TUM Traffic Intersection datasets using BEVFusion and PV-RCNN models. The results show that ActiveAnno3D can achieve high mAP scores with only 50% of the training data, making it a promising solution for efficient and accurate 3D object detection in autonomous driving applications. The framework is also integrated into the proAnno labeling tool to enable AI-assisted data selection and minimize manual labeling costs.