Fig. 1. End-to-end co-design framework for optics and image processing

Fig. 2. Rotationally symmetric PSF and reduces the computational complexity^［6］

Fig. 3. Imaging model based on differentiable ray tracing^［22］

Fig. 4. Typical image reconstruction network， Res-Unet^［6］， Neural-feature deconvolution network^［10］， MultiWienerNet^［24-25］

Fig. 5. Generating error functions using global and local generative adversarial network^［22］

Fig. 6. End-to-end co-design singlet lens^［5］ and compound-lens^［22］ optical systems with large depth-of-field imaging

Fig. 7. End-to-end co-design phase plate achieving 4x super-resolution imaging of SPAD camera^［9］

Fig. 8. End-to-end co-design snapshot computational spectral imaging^［8］

Fig. 9. End-to-end co-design full-spectrum imaging with diffractive lens^［6］

Fig. 10. End-to-end co-design for large field of view imaging^［22］