Quickstart

This guide walks through the core DeepLens workflow: loading a lens, computing PSFs, and rendering images.

Load a Lens

GeoLens is the primary lens model — a differentiable multi-element refractive lens loaded from a JSON, Zemax .zmx, or Code V .seq file.

from deeplens import GeoLens

lens = GeoLens(filename="datasets/lenses/cellphone/cellphone80deg.json")
lens.summary()

Compute a PSF

The point spread function (PSF) describes how the lens images a point source at a given field angle and wavelength.

# Single on-axis PSF (monochromatic)
psf = lens.psf(point=[0.0, 0.0, -10000.0], ks=128, wvln=0.589)

# RGB PSF (weighted sum over visible wavelengths)
psf_rgb = lens.psf_rgb(point=[0.0, 0.0, -10000.0], ks=128)

Visualize the PSF

import matplotlib.pyplot as plt

plt.imshow(psf_rgb.squeeze().permute(1, 2, 0).detach().cpu().numpy())
plt.title("RGB PSF")
plt.axis("off")
plt.show()

Render an Image

Render a full image by convolving with the lens PSF:

import torchvision

img = torchvision.io.read_image("datasets/images/example.png").float() / 255.0
img = img.unsqueeze(0)  # (1, 3, H, W)

rendered = lens.render(img, depth=10000.0)

Lens Types

DeepLens provides several lens models for different use cases:

Lens Type	Description	Use Case
GeoLens	Multi-element refractive ray tracing	Automated lens design, image simulation
HybridLens	Refractive lens + diffractive optical element	Hybrid optics co-design
DiffractiveLens	Pure wave-optics diffractive surfaces	Flat optics, DOE design
PSFNetLens	Neural network PSF surrogate	Fast PSF approximation
ParaxialLens	Thin-lens / circle-of-confusion model	Simple bokeh simulation

Camera Pipeline

Combine a lens with a sensor for end-to-end image simulation:

from deeplens import GeoLens, Camera
from deeplens.sensor import RGBSensor

lens = GeoLens(filename="datasets/lenses/cellphone/cellphone80deg.json")
sensor = RGBSensor(res=(1920, 1080))
camera = Camera(lens=lens, sensor=sensor)

Next Steps

• API Reference — full documentation for all classes
• Examples — lens design, end-to-end optimization, and more