Deep Learning for Single-View Instance Recognition

摘要: Deep learning methods have typically been trained on large datasets in which many training examples are available. However, real-world product only a small number of images available for each product. We explore the use deep recognizing object instances when we single example per class. show that feedforward neural networks outperform state-of-the-art objects from novel viewpoints even just image object. To further improve our performance this task, propose to take advantage supplementary dataset observe separate set multiple viewpoints. introduce new approach instance recognition with limited data, an auxiliary multi-view train network be robust viewpoint changes. find leads more classifier viewpoints, outperforming previous approaches including keypoint-matching, template-based techniques, and sparse coding.