DeepLens
Differentiable optical lens simulator for automated and end-to-end optical design.
DeepLens is a PyTorch-based differentiable simulator for optical systems. It provides gradient-based optimization of lens surfaces, diffractive optical elements, and neural PSF surrogates, and serves as the differentiable optics engine for end-to-end camera pipelines such as End2endImaging.
Key Features
- Differentiable ray tracing through multi-element lens systems with automatic differentiation
- Multiple lens models: geometric (
GeoLens), hybrid refractive-diffractive (HybridLens), pure diffractive (DiffractiveLens), neural surrogate (PSFNetLens), and thin-lens (ParaxialLens) - Accurate image simulation via direct ray tracing, distortion-aware PSF-map rendering, depth-interpolated PSFs, and per-pixel PSF splatting
- Standard lens file I/O: read/write Zemax
.zmx, Code V.seq, and JSON formats - Hybrid ray-wave simulation for JSON-defined refractive lenses with DOE/metasurface phase elements
Quick Install
Getting Started
- Installation — detailed setup instructions
- Quickstart — load a lens, compute a PSF, render an image
- API Reference — full class and function documentation
- Examples — lens design, end-to-end optimization, image simulation