This document outlines the file structure of the deeplens package.

• __init__.py: Package entry point. Exports init_device() (device initialization) and re-exports from subpackages.

• camera.py: Defines a camera system combining a lens and a sensor.

• utils.py: General-purpose utilities (image I/O, batch metrics like PSNR/SSIM, normalization, video creation, logging, seeding).

The optics package is the core module for optical simulations. It hosts all lens types and optical primitives.

• Lens classes (all in optics/):

  • lens.py: Base class Lens for all lens systems.
  • geolens.py: GeoLens — refractive lens systems (ray tracing).
  • diffraclens.py: DiffractiveLens — paraxial diffractive lens systems.
  • hybridlens.py: HybridLens — hybrid refractive-diffractive systems.
  • paraxiallens.py: ParaxialLens — paraxial (thin lens) model for defocus.
  • psfnetlens.py: PSFNetLens — neural surrogate for PSF prediction.

• config.py: Optics configuration constants (DEPTH, SPP_*, PSF_KS, WAVE_RGB, EPSILON, etc.).

• utils.py: Optics utilities (interp1d, grid_sample_xy, foc_dist_balanced, wave_rgb, diff_float, diff_quantize).

• light/: Ray tracing and wave optics.

  • ray.py: Ray class for geometric ray tracing.
  • wave.py: ComplexWave and propagation methods (ASM, etc.).

• material/: Material properties and dispersion models.

• geometric_surface/: Geometric surfaces for refractive lenses (Aspheric, Spheric, Aperture, Plane, etc.).

• diffractive_surface/: Diffractive optical elements and metasurfaces.

• phase_surface/: Phase-only surfaces (Zernike, Fresnel, etc.).

• imgsim/: Image simulation (Monte Carlo, PSF convolution, depth interpolation).

• geolens_pkg/: GeoLens helpers (I/O, optimization, evaluation, tolerance, visualization).

• loss.py: PSF-related loss functions for optical optimization.

Simulates different sensor types and includes an ISP pipeline.

• isp_modules/: ISP modules (demosaic, white balance, gamma, etc.).
• mono_sensor.py, rgb_sensor.py, event_sensor.py: Sensor implementations.

Neural network models.

• surrogate/: Surrogate models (SIREN, MLP, PSFNet).
• reconstruction/: Image reconstruction networks (UNet, NAFNet, Restormer, SwinIR).
• loss/: Training loss functions.
• dataset.py: Dataset utilities.