Dynamic Fusion Network for Light Field Depth Estimation

Focus-based methods have shown promising results for the task of depth estimation in recent years. However, most existing focus-based approaches depend on maximal sharpness focal stack. These ignore spatial relationship between slices. The problem information loss caused by out-of-focus areas stack poses challenges this task. In paper, we propose a dynamically multi-modal learning strategy which incorporates RGB data and our framework. Our goal is to deeply excavate correlation designing pyramid ConvGRU fuse adaptive way dynamic fusion module. success method demonstrated achieving state-of-the-art performance two light field datasets.