The document proposes an object detection method that combines the effectiveness of discriminative detectors with the explicit correspondence of nearest-neighbor approaches. It trains a separate linear SVM classifier for each exemplar in the training set, where each detector is defined by a single positive instance and millions of negatives. While each detector is specific to its exemplar, the paper finds that an ensemble of these Exemplar-SVMs offers good generalization. Their method achieves performance on par with more complex latent part-based models, with only a modest increase in computational cost.