In the digitized era, computer vision technology gained high attention in effectively rendering photo-realistic virtual worlds. Automatic synthesis of photo-realistic images remains an as challenging problem in computer graphics.
The tremendous success of computer vision and machine learning recently provided a new approach to photo-realistic image synthesis and editing. Neural rendering is a novel and quickly efflorescing research field that integrates generative machine learning approaches with physical knowledge from computer graphics.
Some essential cases of neural rendering are novel view synthesis, semantic photo manipulation, facial and body reenactment, relighting, free-viewpoint video, and the creation of photo-realistic avatars for virtual and augmented reality telepresence.
More recently, deep generative models have been widely utilized for automatically generating photo-realistic images and videos with neural rendering. An outstanding neural rendering model achieves a controllable, realistic model of a 3D scene.
Neural rendering is a leading-edge technology for synthesizing realistic image and video data. Digital Assistants, De-aging or Digital Beauty work, Video Dialogue Replacement, and Face swap impersonation are some advanced neural rendering implementations. Various neural rendering applications are highlighted below;
Semantic Photo Synthesis and Manipulation – Semantic Photo Synthesis and Semantic Image Manipulation improve the Realism of Synthetic Renderings are helps to control and modify the appearance of a photograph in a semantically purposeful way.
Novel View Synthesis for Objects and Scenes – Novel view synthesis methods are Neural Image-based Rendering, Neural Rerendering, Novel View Synthesis with Multi-plane Images, Neural Scene Representation and Rendering, Voxel-based Novel View Synthesis Methods, and Voxel-based Novel View Synthesis Methods.
Free Viewpoint Videos – Volumetric performance capture is designed to acquire the 3D shape and texture, and its recent developments are LookinGood with Neural Rerendering, Neural Volumes, and Free Viewpoint Videos from a Single Sensor.
Learning to Relight – Recently developed relighting methods are Deep Image-based Relighting from Sparse Samples, Multi-view Scene Relighting, Deep Reflectance Fields, and Single Image Portrait Relighting.
Facial Reenactment – Facial reenactment is applied to alter the scene properties beyond the viewpoints. Some of their currently developed systems are Single Image Portrait Relighting, Editing Video by Editing Text, Image Synthesis using Neural Textures, Neural Talking Head Models, and Deep Appearance Models.
Body Reenactment – Body reenactment is developed to resolve the problem of synthesizing realistic images of the full human body. Full body performance cloning approaches, Textured Neural Avatars, and the DensePose model are currently evolved approaches for body reenactment