Real-time Car Part Instance Segmentation: the Comparison of the State-of-the-Art

This study investigates Deep Learning’s (DL) impact on car part segmentation, particularly through Convolutional Neural Networks (CNNs). While frameworks like Mask R-CNN improve object detection, efficiency concerns motivate the exploration of faster models. Our research prioritizes high-precision car part segmentation for real-world applications. We compared one-stage and multi-stage models, selecting Mask R-CNN, DetectoRS, QueryInst, RTMDet, and YOLOv8 for car part segmentation. Utilizing the 'Humans in the Loop' dataset with 998 extensively annotated car part images, the study contributes to computer vision advancement in the automotive domain. Training the selected models with on-the-fly data augmentation and evaluated using mean Average Precision (mAP) and mAP-50 metrics for bounding box evaluation. YOLOv8, a real-time one-stage model, achieved mAP scores of 0.70 and mAP50 scores of 0.89, highlighting its potential for efficient in car part segmentation task.