Image rendering via neural radiance fields

Neural radiance fields (NeRF) can learn a 5-dimensional image model to be employed for rendering scenes from arbitrary viewpoints, employing neural networks that are trained to fit a specific scene. These models have several applications in image processing and vision.

We are developing NERF models to render an image at a higher resolution than that of the available views, exploiting the correlation among the views themselves.

We are also working towards the development of multimodal NeRF systems that can take advantage of different imaging modalities (image and depth, image and point cloud, ...) in order to render the scene with the best possible quality.

Link:

ERC Sector:

  • PE7_7 Signal processing

Keywords:

  • Artificial neural networks
  • Image processing
  • Deep learning
  • Image restoration

Persone di riferimento