API Reference

End2endImaging works as a computational camera — a differentiable model of the full image-capture pipeline. A computational camera is composed of three components: an optical lens (provided by DeepLens), an image sensor, and a reconstruction network. The Camera ties the lens and sensor into a single differentiable capture model, which pairs with a network for end-to-end design.

• Camera

  ---

  The computational camera that ties an optical lens and sensor into one differentiable capture for image simulation and end-to-end co-design.

  Camera

• Optics (DeepLens)

  ---

  Differentiable lens models, optical surfaces, and PSF simulation — provided by the DeepLens optics engine.

  Optics (DeepLens)

• Sensor

  ---

  Differentiable RGB/mono sensors with read/shot noise and a full ISP pipeline (demosaic, white balance, color correction, gamma).

  Sensor

• Network

  ---

  Image-reconstruction networks (UNet, Restormer, NAFNet) and training losses (PSNR, SSIM, perceptual).

  Network

The optical lens models (GeoLens, HybridLens, DiffractiveLens, …) are documented in the DeepLens API reference.