Janssen is a JAX-based library for optical microscopy simulations and ptychographic reconstructions. It provides differentiable implementations of optical propagation, lens simulations, and phase retrieval algorithms, all optimized for GPU acceleration through JAX.
Janssen follows a layered architecture built on JAX's functional paradigm:
- Foundation Layer: Core data structures (
types) as JAX PyTrees for automatic differentiation, plus math utilities (utils) including Bessel functions and Wirtinger gradients - Physics Core: Optical propagation (
prop), lens models (lenses), beam profiles (models), and wavefront manipulation (optics) - Coherence Layer: Partial coherence simulation (
cohere) with spatial/temporal modes and source models - Application Layer: Forward microscope models (
scopes), phase retrieval algorithms (invert), and visualization (plots)
Every optical element is a pure function Wavefront → Wavefront, enabling JAX to trace computation graphs and compute gradients automatically for inverse problems like ptychography.