Attention Mechanism-Based Light-Field View Synthesis

The angular information of light lost in conventional images but preserved and stored light-fields plays an instrumental role many applications such as depth estimation, 3D reconstruction post-capture refocusing. However, the limited resolution light-field due to consumer hardware limitations is a major drawback its widespread adoption. In this article, we present novel deep learning-based view synthesis method from sparse set input views. Our proposed method, end-to-end trainable network, utilizes convolutional block attention modules enhance built-in image-based rendering. module consists two sequentially applied channel spatial dimensions focus network on critical features. architecture combination three sub-networks, one for stereo feature extraction, another disparity last attention-based refinement. We schemes refinement that perform equally well differ number parameters by 1.2% execution time 44%. Quantitative qualitative results four challenging real-world datasets show superiority method. shows considerable PSNR gains, around 1 dB compared state art 0.5 over our previous LFVS-AM. addition, ablation studies demonstrate effectiveness each Finally, parallax edge precision-recall curve better preserves details.