Deep learning-based viewpoint recommendation in volume visualization

Viewpoint is vital in guiding the user to understand the volume data. However, a model that can recommend viewpoints conforming to user preference is hard to be represented explicitly. In this work, we propose an implicit model for the best viewpoint recommendation of volume visualization with CNN-based models to learn the traditional scoring method and user preference. Residual structures are applied for reducing overfitting in simple scalar regression and solving the problem of accuracy getting lower as the network getting deeper. Multi-level-based structures are applied to imitate the coarse and fine level in human perception. The detailed experiments of comparison between our model and traditional methods confirm the efficiency of our work. A case of application verifies that our model can flexibly realize a user preference-based best viewpoint selection in volume visualization. Graphic abstract