This paper addresses the issue of domain shift in object recognition, a critical topic in computer vision. The authors introduce a novel method that adapts object models from a source domain to a target domain with different imaging conditions by learning a transformation that minimizes the impact of domain-induced changes in the feature distribution. This transformation is learned through supervised learning and can be applied to categories with few or no labeled examples in the target domain. The method is demonstrated to improve recognition performance on categories with moderate to large changes in imaging conditions. The paper also contributes a new multi-domain object database and discusses the importance of domain adaptation in object recognition, especially when labeled datasets are limited or not representative of real-world applications. The proposed technique is based on cross-domain transformations, which map points from the source domain closer to those in the target domain using labeled pairs from both domains. This approach can be applied to novel test samples and generalizes to new categories not seen during training.This paper addresses the issue of domain shift in object recognition, a critical topic in computer vision. The authors introduce a novel method that adapts object models from a source domain to a target domain with different imaging conditions by learning a transformation that minimizes the impact of domain-induced changes in the feature distribution. This transformation is learned through supervised learning and can be applied to categories with few or no labeled examples in the target domain. The method is demonstrated to improve recognition performance on categories with moderate to large changes in imaging conditions. The paper also contributes a new multi-domain object database and discusses the importance of domain adaptation in object recognition, especially when labeled datasets are limited or not representative of real-world applications. The proposed technique is based on cross-domain transformations, which map points from the source domain closer to those in the target domain using labeled pairs from both domains. This approach can be applied to novel test samples and generalizes to new categories not seen during training.