Skip to content

Usage 🏗️💻

The methodology of Pyro-NN is contained within the ct_reconstruction folder, which is organized into four essential parts:

1. 🔧 Geometry

Defines the scanning parameters and trajectory.

  • Initialization from parameters: This is possible if you know all your scanning parameters and if the scanning trajectory was circular.
  • ⚠️ Small Tip: Sometimes, parameters in the header can be incorrectly filled. Be aware—errors may still occur!

2. 🛠️ Layers

Defines the 2D and 3D forward/backward projectors.

  • Geometry setup: To initialize these layers, the geometry of the scan must be defined.

  • Input of all layers: The input is the Image-Tensor (depending on the dimensionality) and a geometry dictionary, returned when the geometry is initialized.

  • For 2D: Includes implementations for Parallel Beam and Fan Beam.

  • For 3D: Implements Cone Beam.

3. 🔍 Helpers

Provides pre-implemented filters, weights, trajectories, and phantoms.

  • Implemented Filters:
  • Ramp Filters, Ram Lak, Shepp Logan, Cosine, Hamming, Hann 🎛️
  • Implemented Weights:
  • Cosine, Parker ⚖️
  • Implemented Trajectories:
  • Circular and arbitrary paths 🌀

4. ⚙️ Cores

Contains the kernels and the PyTorch connection.

        Each part plays a crucial role in making Pyro-NN an efficient 
        and powerful framework for differentiable reconstruction. 🚀