This paper introduces a novel method called Ensemble of Exemplar-SVMs (ESVMs) for object detection, which combines the effectiveness of discriminative object detectors with the explicit correspondence provided by a nearest-neighbor approach. Each ESVM is trained on a single positive example and millions of negative examples, making it highly specific to the exemplar it is trained on. The ensemble of these ESVMs offers good generalization, performing on par with more complex latent part-based models while requiring only a modest increase in computational cost. The key benefit of ESVMs is that they create an explicit association between each detection and a single training exemplar, allowing for the transfer of metadata such as segmentation, geometry, and 3D models directly to the detections. This enables a wide range of applications beyond object detection, including scene understanding and computer graphics. The paper also discusses the motivation behind the approach, the method's architecture, and its implementation details, followed by experimental evaluations on the PASCAL VOC 2007 dataset and other tasks like segmentation, geometry estimation, 3D model transfer, and related object priming.This paper introduces a novel method called Ensemble of Exemplar-SVMs (ESVMs) for object detection, which combines the effectiveness of discriminative object detectors with the explicit correspondence provided by a nearest-neighbor approach. Each ESVM is trained on a single positive example and millions of negative examples, making it highly specific to the exemplar it is trained on. The ensemble of these ESVMs offers good generalization, performing on par with more complex latent part-based models while requiring only a modest increase in computational cost. The key benefit of ESVMs is that they create an explicit association between each detection and a single training exemplar, allowing for the transfer of metadata such as segmentation, geometry, and 3D models directly to the detections. This enables a wide range of applications beyond object detection, including scene understanding and computer graphics. The paper also discusses the motivation behind the approach, the method's architecture, and its implementation details, followed by experimental evaluations on the PASCAL VOC 2007 dataset and other tasks like segmentation, geometry estimation, 3D model transfer, and related object priming.