Flexible SVBRDF Capture with a Multi-Image Deep Network

Valentin Deschaintre, Miika Aittala, Fredo Durand, George Drettakis and Adrien Bousseau

Data

• Materials dataset (1.1GB)

Code and trained weights.

• Method's code (4MB)
• Trained weights (800 MB)

Supplemental materials.

Results

• Our method on many real data, from 1 to 5 input images
• Our method on data acquired with only one phone, from 1 to 5 input images
• RMSE of our predictions with respect to the number of input images, averaged over our synthetic test dataset

Comparisons

• Comparison to Deschaintre et al.18 and Li et al.18 on real Data
• Comparison to Deschaintre et al.18 and Li et al.18 on synthetic Data
• RMSE on re-renderings for the maps obtained by our method with 5 images (dotted blue) and by a classical optimization method with an increasing number of input images (black)