One embodiment provides a method and system for performing segmentation on vehicle damage images. During operation, the system obtains a plurality of labeled segmentation and target detection samples. A labeled segmentation sample includes an image of a target object, a classification of the target object, and an outline of the target object; and a labeled target detection sample includes an image of a target object and a classification of the target object. The system trains a segmentation model, which comprises determining, based on labeled segmentation samples, mapping between a target detection parameter associated with target objects of a classification and a target segmentation parameter associated with outlines of the target objects; determining a target detection parameter for each classification; and obtaining a target segmentation parameter for each classification based on the corresponding target detection parameter and mapping, thereby determining the segmentation model for performing segmentation on vehicle damage images.