Neural network based predictors for 3D content streaming and rendering

3D content streaming and rendering system has attracted a significant attention from both academia and industry. However, these systems struggle to provide comparable quality to that of locally stored and rendered 3D data. Since the rendered 3D content on to the client machine is controlled by the users, their interactions have a strong impact on the performance of 3D content streaming and rendering system. Thus, considering user behaviors in these systems could bring significant performance improvements. In this paper, an Artificial Neural Network (ANN) based predictor is proposed for 3D content streaming and rendering. The user interactions on various 3D contents are profiled and used as information to train the Neural Network predictors. The 3D content could be static or dynamic 3D object / scene. We test our model through another set of interactions over the 3D contents by same users. The tested result shows that our model can learn the user interactions and is able to predict several interactions to help in optimizing the streaming and rendering for better performance. We also propose various approaches based on traces collected from the same/different users to accelerate the learning process of the neural network.