The one-shot contour primitive of interest (CPI) extraction method is meaningful for flexible machine vision, which can obtain the target CPI on a novel object under the guidance of a support image. The one-shot guidance is based on an object-agnostic feature extractor and the feature comparison in a metric space. Therefore, the discrimination and generalization abilities of the object-agnostic feature extractor are essential to the one-shot CPI extraction. In this paper, we propose an object-specific rapid feature adaption method to enhance a trained CPI extraction network. Three lightweight feature adaption modules are designed to adjust the features specifically for the present object type. A module can be flexibly plugged into a trained network at three different candidate positions, to influence the network’s final outputs. The parameters of the module are learned with the annotated support sample and only tens of optimization steps in several seconds. Thus, although the feature adaption module only works for a specific object, it can be deployed flexibly and rapidly. The effectiveness of the proposed method is validated with a series of experiments.