Filter-Based Reconstruction for
Gradient-Domain Rendering



Difei Yan, Shaokun Zheng, Ling-Qi Yan, and Kun Xu


Comparisons

This interactive viewer compares our reconstructed results with the following methods: L1 reconstruction (L1), L2 reconstruction (L2), GradNet (GradNet), and Deep Convolutional Reconstruction for Gradient-Domain Rendering (NGPT). All input images are sampled with the same sample rates. The reference image is rendered under 524288 samples per pixel (spp).

(1) 64 spp

Ours
CV

(2) 256 spp

Ours
CV