DaFISH
A lighting disentanglement method for novel view synthesis that represents illumination using spherical harmonics, decomposing each point's emitted color into surface albedo and SH lighting factors. It can be inserted into any NVS pipeline to address view-dependent lighting problems in scenarios where active light sources are rigidly coupled to the camera.
Timeline (last 90 days)
DaFISH is a lighting disentanglement method that can be plugged into any novel view synthesis pipeline, representing lighting via spherical harmonics and decomposing each point's emitted color into surface albedo and an SH lighting factor.
On held-out frames across five scenes, DaFISH achieved the highest average PSNR (38.91 dB), SSIM (0.9613), and LPIPS (0.0582), outperforming the strongest splatting baseline and the closest NeRF baseline.
Once trained, DaFISH can compute lighting analytically for any camera pose without retraining, thereby generalizing to unseen viewpoints.
The key to DaFISH is that the SH-to-irradiance conversion has a closed-form solution and the integral in volume rendering is differentiable, so the albedo, SH lighting coefficients, and geometry parameters can be updated simultaneously via backpropagation.
DaFISH integrates seamlessly with existing NVS pipelines (such as NeRF), verifying its generality and effectiveness.
DaFISH currently targets mainly static scenes, and extension to dynamic scenes or complex lighting changes (such as multiple moving light sources) still requires further research.
All Facts (6)
DaFISH is a lighting disentanglement method that can be plugged into any novel view synthesis pipeline, representing lighting via spherical harmonics and decomposing each point's emitted color into surface albedo and an SH lighting factor.
50%UnverifiedOn held-out frames across five scenes, DaFISH achieved the highest average PSNR (38.91 dB), SSIM (0.9613), and LPIPS (0.0582), outperforming the strongest splatting baseline and the closest NeRF baseline.
50%UnverifiedOnce trained, DaFISH can compute lighting analytically for any camera pose without retraining, thereby generalizing to unseen viewpoints.
50%UnverifiedThe key to DaFISH is that the SH-to-irradiance conversion has a closed-form solution and the integral in volume rendering is differentiable, so the albedo, SH lighting coefficients, and geometry parameters can be updated simultaneously via backpropagation.
50%UnverifiedDaFISH integrates seamlessly with existing NVS pipelines (such as NeRF), verifying its generality and effectiveness.
50%UnverifiedDaFISH currently targets mainly static scenes, and extension to dynamic scenes or complex lighting changes (such as multiple moving light sources) still requires further research.
50%