A Lightweight Depth Estimation Network for Wide-Baseline Light Fields

Existing traditional and ConvNet-based methods for light field depth estimation mainly work on the narrow-baseline scenario. This paper explores feasibility capability of ConvNets to estimate in another promising scenario: wide-baseline fields. Due deficiency training samples, a large-scale diverse synthetic dataset with labelled data is introduced prediction tasks. Considering practical goal real-world applications, we design an end-to-end trained lightweight convolutional network infer depths from fields, called LLF-Net. The proposed LLF-Net built by incorporating cost volume which allows variable angular inputs attention module that enables recover details at occlusion areas. Evaluations are made experimental results show achieves best performance when compared recent state-of-the-art methods. We also evaluate our datasets, it consequently improves previous