Abstract:
Long-term image restoration and manipulation goals include learning a good image prior. Existing methods like deep image prior (DIP) capture low-level image statistics, but there are still gaps toward capturing rich image semantics like color, spatial coherence, textures, and high-level concepts. A generative adversarial network (GAN) trained on large-scale natural images can be used to exploit the image prior in this work. As shown in Fig. 1, the deep generative prior (DGP) restores color, patch, and resolution of degraded images. It allows random jittering, image morphing, and category transfer. Relaxing the generator-fixing assumption of GAN inversion methods allows for highly flexible restoration and manipulation. We allow the generator to be fine-tuned on-the-fly in a progressive manner regularized by GAN discriminator feature distance. These simple and practical changes preserve the reconstruction in the manifold of nature images, resulting in more precise and faithful reconstruction for real images.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
