FEAMNet: Light Field Depth Estimation Network Based On Feature Extraction and Attention Mechanism
Yunming Liu, Yuxuan Pan, Kaiyue Luo, Yu Liu, Lin Zhang · 2023
With the rich 4D visual information of light rays, light field (LF) applications contribute to the development of immersive multimedia and virtual reality, where LF depth estimation is the critical problem. However, existing light field depth estimation algorithms deliver weak performance in edge and textureless regions. In this paper, we propose the feature-extraction and attention mechanism-based network (FEAMNet), which can effectively handle edges and textureless regions in depth maps. The FEAMNet contains a dilated-convolution and average-pooling (DCAP) feature extraction module with a large receptive field to acquire multi-scale features. And the channel attention-based disparity regression (CADR) module is introduced to measure the importance weights of different feature channels for high accuracy. The experimental results show that the FEAMNet outperforms state-of-the-art algorithms, such as OACC and DistgDisp. The implementation of our FEAMNet with the mentioned dataset is open sourced at https://github.com/lymwxq/FEAMNet.