Dual Stage Semantic Information Based Generative Adversarial Network for Image Super-Resolution

Generative adversarial network
DOI: 10.2139/ssrn.4227519 Publication Date: 2022-09-24T09:17:43Z
ABSTRACT
Deep learning methods for the super-resolution problem are showing great performance compared to other traditional techniques. However, these unable learn complex spatial structures and high frequency details; which leads over-smooth results. In present paper, a novel Generative Adversarial Network based architecture named as Residue Semantic feature Dual Subpixel has been proposed generator discriminator networks solve problem. The network is residue semantic dual subpixel generative architecture. This divided into two stages: premier residual stage deuxieme stage. These stages concatenated together form stageupsamping process, enhances capability of our model. Inter intra connections made withinthese stages; helping us sustain texture details images. information implanted in enhance quality objects an image. For embedding generator, maps extracted from pre-trained model merged with input To stabilize training we introduced spectral normalization discriminator. Visual perception mean opinion score shows that method outperforms state-of-the-art methods.
SUPPLEMENTAL MATERIAL
Coming soon ....
REFERENCES (0)
CITATIONS (0)