Training deep learning model requires massive datasets, but data collection and labeling can be highly time-consuming and labor-intensive. To address this, we developed an Auto-Labeling technology that can automatically labeling both 2D and 3D data.

The developed Auto-Labeling technology based on an object detection model that integrates 2D images and 3D point cloud to automatically label 2D and 3D bounding box.This process not only significantly improves labeling speed, accuracy but also greatly reduces labor cost.

Our Auto-Labeling technology achieves over 90% precision and 90% recall rates. As shown in the experimental results, the labeling performance of Auto-Labeling is comparable to manual annotation and can even detect a few objects missed by human.This technology is a powerful tool toassist labeling and effectively reduce manual labor cost by approximately 80% to 90%.

Result