Prototype Queue Learning For Multi-Class Few-Shot Semantic Segmentation

A regular photographic picture provides 2-dimensional information of the object by recording only intensity of light, whereas a CGH provides 3-dimensional information about the object, including distance information, by recording light interference. CGH is becoming possible to display high-resolution CGH images due to advance in display technology. However, CGH requires a great deal of time to generate a high-resolution image because it calculates large volumes of information, such as light reflected from an object and light interference. To solve this problem, we propose a model that predicts interference information and generates high-resolution CGH images from low-resolution CGH images using a deep learning model. We propose a dual-generator GAN model consisting of two generators and one discriminator, and compare the results with existing models that generate high-resolution CGH images. The generated high-resolution CGH images were measured and evaluated using a SSIM and PSNR indicators.