Viewpoint Estimation for Objects with Convolutional Neural Network Trained on Synthetic Images

In this paper, we propose a method to estimate object viewpoint from a single RGB image and address two problems in estimation: generating training data with viewpoint annotations and extracting powerful features for the estimation. We first collect 1780 high quality 3D CAD object models of 3 categories. Then we generate a synthetic RGB image dataset with viewpoint annotations, in which each image is generated by placing one model in a realistic panorama scene and rendering the model with a random camera parameters. We train a CNN model on our synthetic dataset to predict the object viewpoint. The proposed method is evaluated on PASCAL 3D+ dataset and our synthetic dataset. The experiment results show good performance.