MaterialGAN: Reflectance Capture using a Generative SVBRDF Model

Supplemental Materials

Yu Guo1, Cameron Smith2, Miloš Hašan2, Kalyan Sunkavalli2, and Shuang Zhao1
1University of California, Irvine          2Adobe Research
Real Data (Detailed Comparisons)
We show detailed comparisons of Gao19 and our technique using real measurements in the following. Methods marked as "+" are initialized with Deschaintre19's direct predictions; those without "+" are obtained using the better of two constant initializations. We also show the result with and without post-refinement, for both our and Gao19 results. The lowest and second lowest LPIPS values in each column are marked in green and orange, respectively.

plastic-red-carton

Remark: In this example, all versions of our method (with or without initialization or refinement) give a fairly clean result, while all versions of Gao19 produce artifacts in roughness and specular maps. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
no refinement
LPIPS: 0.364
Ours
no refinement
LPIPS: 0.123
[Gao19]
LPIPS: 0.187
Ours
LPIPS: 0.095
[Gao19]+
no refinement
LPIPS: 0.229
Ours+
no refinement
LPIPS: 0.115
[Gao19]+
LPIPS: 0.166
Ours+
LPIPS: 0.095

cards-red

Remark: On this Christmas card example with colorful specular coefficient variation, both methods perform reasonably well. Our method has somewhat better LPIPS scores, with or without initialization. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
no refinement
LPIPS: 0.377
Ours
no refinement
LPIPS: 0.073
[Gao19]
LPIPS: 0.096
Ours
LPIPS: 0.059
[Gao19]+
no refinement
LPIPS: 0.310
Ours+
no refinement
LPIPS: 0.103
[Gao19]+
LPIPS: 0.092
Ours+
LPIPS: 0.061

wood-tile

Remark: In this example, all versions of our method (with or without initialization or refinement) give a fairly clean result, while all versions of Gao19 produce artifacts in roughness and specular maps. While the LPIPS scores for Gao19 with refinement are good, the maps are clearly worse than ours. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
no refinement
LPIPS: 0.600
Ours
no refinement
LPIPS: 0.295
[Gao19]
LPIPS: 0.193
Ours
LPIPS: 0.202
[Gao19]+
no refinement
LPIPS: 0.489
Ours+
no refinement
LPIPS: 0.204
[Gao19]+
LPIPS: 0.192
Ours+
LPIPS: 0.164

book1

Remark: This example is difficult for all methods, due to sharply segmented materials and layered BRDFs that do not match the simple microfacet model well. None of the methods can closely match the novel lighting. Our method without initialization produces cleanly segmented maps and also correctly estimates the normal as mostly flat; its LPIPS score is also best. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
no refinement
LPIPS: 0.396
Ours
no refinement
LPIPS: 0.213
[Gao19]
LPIPS: 0.264
Ours
LPIPS: 0.147
[Gao19]+
no refinement
LPIPS: 0.418
Ours+
no refinement
LPIPS: 0.286
[Gao19]+
LPIPS: 0.318
Ours+
LPIPS: 0.235
Real Data
We now show real examples generated using Gao19 and our technique. Methods marked as "+" are initialized with Deschaintre19's direct predictions; those without "+" are obtained using the better of two constant initializations. The lowest and second lowest LPIPS values in each column are marked in green and orange, respectively.

wall-plaster-green

Remark: Our results, with or without initialization by Deschaintre19, have significantly reduced artifacts compared to Gao19, in roughness and specular maps. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.315
Ours
LPIPS: 0.199
[Gao19]+
LPIPS: 0.271
Ours+
LPIPS: 0.187

wall-plaster-white

Remark: Our results, with or without initialization by Deschaintre19, have significantly reduced artifacts compared to Gao19, in roughness and specular maps. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.238
Ours
LPIPS: 0.071
[Gao19]+
LPIPS: 0.132
Ours+
LPIPS: 0.089

wall-red-bump

Remark: Our results, with or without initialization by Deschaintre19, have significantly reduced artifacts compared to Gao19, in roughness and specular maps. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.090
Ours
LPIPS: 0.060
[Gao19]+
LPIPS: 0.090
Ours+
LPIPS: 0.058

plastic-foam

Remark: All methods perform similarly in map quality. Our methods have lowest LPIPS scores.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.132
Ours
LPIPS: 0.076
[Gao19]+
LPIPS: 0.123
Ours+
LPIPS: 0.060

plastic-redgrid

Remark: Our results, with or without initialization by Deschaintre19, have significantly reduced artifacts compared to Gao19, in roughness and specular maps. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.100
Ours
LPIPS: 0.095
[Gao19]+
LPIPS: 0.103
Ours+
LPIPS: 0.116

leather-blue

Remark: The uninitialized methods are computing more correct roughness maps. Roughness initialization from Deschaintre19 is too low here, and contains an artifact which Gao's method fails to remove. Ours has sharper maps and lower LPIPS than [Gao19].
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.223
Ours
LPIPS: 0.146
[Gao19]+
LPIPS: 0.356
Ours+
LPIPS: 0.315

leather-darkbrown

Remark: Both optimization methods (ours and Gao19) perform similarly, with or without initialization. Our LPIPS scores are slightly better. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.288
Ours
LPIPS: 0.225
[Gao19]+
LPIPS: 0.282
Ours+
LPIPS: 0.205

bathroomtile1

Remark: Our maps (with or without initialization) have fewer artifacts than Gao19, especially in specular map. For re-rendering on novel views, none of the methods are very close. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.264
Ours
LPIPS: 0.186
[Gao19]+
LPIPS: 0.222
Ours+
LPIPS: 0.235

bathroomtile2

Remark: Our results, with or without initialization by Deschaintre19, have significantly reduced artifacts compared to Gao19, in roughness and specular maps. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.242
Ours
LPIPS: 0.225
[Gao19]+
LPIPS: 0.231
Ours+
LPIPS: 0.192

wood-jatoba

Remark: Our results, with or without initialization by Deschaintre19, have significantly reduced artifacts compared to Gao19, in roughness and specular maps. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.225
Ours
LPIPS: 0.252
[Gao19]+
LPIPS: 0.210
Ours+
LPIPS: 0.225

wood-walnut

Remark: Our results, with or without initialization by Deschaintre19, have significantly reduced artifacts compared to Gao19, in roughness and specular maps. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.283
Ours
LPIPS: 0.226
[Gao19]+
LPIPS: 0.252
Ours+
LPIPS: 0.199

wood-t

Remark: Despite the slightly better LPIPS score of [Gao19]+ over Ours, both of our results have fewer artifacts in the maps.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.173
Ours
LPIPS: 0.199
[Gao19]+
LPIPS: 0.167
Ours+
LPIPS: 0.147

wood-alder

Remark: Our results, with or without initialization by Deschaintre19, have significantly reduced artifacts compared to Gao19, in roughness and specular maps. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.240
Ours
LPIPS: 0.174
[Gao19]+
LPIPS: 0.183
Ours+
LPIPS: 0.176

wood-knotty

Remark: Our results, with or without initialization by Deschaintre19, have significantly reduced artifacts compared to Gao19, in roughness and specular maps. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.234
Ours
LPIPS: 0.164
[Gao19]+
LPIPS: 0.200
Ours+
LPIPS: 0.171

wood-bamboo

Remark: All methods have some amount of burn-in on this highly specular material. Still, in Ours+ the artifacts are significantly reduced, and our methods have better LPIPS scores on novel views. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.211
Ours
LPIPS: 0.149
[Gao19]+
LPIPS: 0.200
Ours+
LPIPS: 0.135

other-bamboo-veawe

Remark: Our results, with or without initialization by Deschaintre19, have significantly reduced artifacts compared to Gao19, in roughness and specular maps. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.234
Ours
LPIPS: 0.120
[Gao19]+
LPIPS: 0.183
Ours+
LPIPS: 0.100

stone-bigtile

Remark: Ours+ has the best result, for both map quality and LPIPS score. Gao19 has artifacts in roughness and specular maps. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.138
Ours
LPIPS: 0.140
[Gao19]+
LPIPS: 0.142
Ours+
LPIPS: 0.113

stone-smalltile

Remark: All methods perform well on re-rendering. However, our maps have fewer artifacts than Gao19.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.069
Ours
LPIPS: 0.070
[Gao19]+
LPIPS: 0.076
Ours+
LPIPS: 0.079

stone-bathroom-tile

Remark: Our results, with or without initialization by Deschaintre19, have significantly reduced artifacts compared to Gao19, in roughness and specular maps. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.231
Ours
LPIPS: 0.197
[Gao19]+
LPIPS: 0.204
Ours+
LPIPS: 0.173

rubber-pattern

Remark: This is a tricky example of a dielectric with close to zero diffuse component, and a highly structured normal. All methods produce somewhat plausible maps, but with various artifacts. Ours has the best roughness and specular map, and lowest LPIPS.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.109
Ours
LPIPS: 0.097
[Gao19]+
LPIPS: 0.114
Ours+
LPIPS: 0.188

book2

Remark: Our results, with or without initialization by Deschaintre19, have significantly reduced artifacts compared to Gao19, in roughness and specular maps. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.160
Ours
LPIPS: 0.042
[Gao19]+
LPIPS: 0.122
Ours+
LPIPS: 0.055

cards-blue

Remark: On this Christmas card example with colorful specular coefficient variation, both methods perform reasonably well. Ours+ has arguably the sharpest and most realistic maps, despite very slightly lower LPIPS. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.078
Ours
LPIPS: 0.075
[Gao19]+
LPIPS: 0.075
Ours+
LPIPS: 0.076

giftbag1

Remark: Another example with colorful specular coefficient variation. Our methods have fewer artifacts, with Ours+ having the cleanest/sharpest maps, despite a slightly lower LPIPS than Ours.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.234
Ours
LPIPS: 0.183
[Gao19]+
LPIPS: 0.218
Ours+
LPIPS: 0.193

giftbag2

Remark: Our results, with or without initialization by Deschaintre19, have significantly reduced artifacts compared to Gao19, in roughness and specular maps. Deschaintre19 produces visually clean maps, but re-rendered appearance is far from the reference.
SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
LPIPS: 0.196
Ours
LPIPS: 0.141
[Gao19]+
LPIPS: 0.168
Ours+
LPIPS: 0.155
Synthetic Data
In what follows, we show synthetic examples generated using Gao19 and our technique. Methods marked as "+" are initialized with Deschaintre19's direct predictions; those without "+" are obtained using the better of two constant initializations. The lowest and second lowest RMSE/LPIPS values in each column are marked in green and orange, respectively. We find that all methods perform relatively well on synthetic data compared to real data, possibly because of the exact BRDF model match and exact calibration, and also because the synthetic examples, while distinct from training examples, are relatively similar in style to the training set. Thus, the differences between the methods tend to be smaller than on real data.

1

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.056RMSE: 0.053RMSE: 0.258RMSE: 0.213 LPIPS: 0.224
Ours
RMSE: 0.060RMSE: 0.060RMSE: 0.301RMSE: 0.138 LPIPS: 0.163
[Gao19]+
RMSE: 0.061RMSE: 0.056RMSE: 0.303RMSE: 0.177 LPIPS: 0.186
Ours+
RMSE: 0.061RMSE: 0.052RMSE: 0.262RMSE: 0.155 LPIPS: 0.158

2

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.038RMSE: 0.034RMSE: 0.201RMSE: 0.092 LPIPS: 0.182
Ours
RMSE: 0.041RMSE: 0.041RMSE: 0.232RMSE: 0.078 LPIPS: 0.182
[Gao19]+
RMSE: 0.042RMSE: 0.031RMSE: 0.247RMSE: 0.072 LPIPS: 0.174
Ours+
RMSE: 0.041RMSE: 0.032RMSE: 0.243RMSE: 0.067 LPIPS: 0.166

3

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.045RMSE: 0.040RMSE: 0.075RMSE: 0.087 LPIPS: 0.054
Ours
RMSE: 0.030RMSE: 0.046RMSE: 0.077RMSE: 0.057 LPIPS: 0.059
[Gao19]+
RMSE: 0.025RMSE: 0.017RMSE: 0.058RMSE: 0.055 LPIPS: 0.030
Ours+
RMSE: 0.023RMSE: 0.027RMSE: 0.048RMSE: 0.036 LPIPS: 0.026

4

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.016RMSE: 0.019RMSE: 0.094RMSE: 0.089 LPIPS: 0.086
Ours
RMSE: 0.016RMSE: 0.016RMSE: 0.067RMSE: 0.048 LPIPS: 0.073
[Gao19]+
RMSE: 0.017RMSE: 0.009RMSE: 0.055RMSE: 0.046 LPIPS: 0.048
Ours+
RMSE: 0.014RMSE: 0.013RMSE: 0.050RMSE: 0.051 LPIPS: 0.062

5

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.055RMSE: 0.034RMSE: 0.271RMSE: 0.197 LPIPS: 0.238
Ours
RMSE: 0.040RMSE: 0.048RMSE: 0.073RMSE: 0.191 LPIPS: 0.204
[Gao19]+
RMSE: 0.045RMSE: 0.022RMSE: 0.157RMSE: 0.196 LPIPS: 0.128
Ours+
RMSE: 0.035RMSE: 0.024RMSE: 0.090RMSE: 0.168 LPIPS: 0.110

6

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.082RMSE: 0.044RMSE: 0.155RMSE: 0.286 LPIPS: 0.233
Ours
RMSE: 0.041RMSE: 0.042RMSE: 0.102RMSE: 0.192 LPIPS: 0.174
[Gao19]+
RMSE: 0.049RMSE: 0.027RMSE: 0.098RMSE: 0.196 LPIPS: 0.104
Ours+
RMSE: 0.042RMSE: 0.025RMSE: 0.085RMSE: 0.173 LPIPS: 0.086

7

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.058RMSE: 0.029RMSE: 0.043RMSE: 0.111 LPIPS: 0.098
Ours
RMSE: 0.047RMSE: 0.018RMSE: 0.020RMSE: 0.071 LPIPS: 0.043
[Gao19]+
RMSE: 0.043RMSE: 0.012RMSE: 0.015RMSE: 0.042 LPIPS: 0.032
Ours+
RMSE: 0.029RMSE: 0.011RMSE: 0.012RMSE: 0.048 LPIPS: 0.012

8

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.134RMSE: 0.020RMSE: 0.146RMSE: 0.145 LPIPS: 0.235
Ours
RMSE: 0.093RMSE: 0.022RMSE: 0.095RMSE: 0.173 LPIPS: 0.081
[Gao19]+
RMSE: 0.161RMSE: 0.025RMSE: 0.064RMSE: 0.122 LPIPS: 0.152
Ours+
RMSE: 0.077RMSE: 0.019RMSE: 0.082RMSE: 0.129 LPIPS: 0.056

9

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.030RMSE: 0.050RMSE: 0.061RMSE: 0.095 LPIPS: 0.126
Ours
RMSE: 0.036RMSE: 0.054RMSE: 0.103RMSE: 0.059 LPIPS: 0.125
[Gao19]+
RMSE: 0.023RMSE: 0.023RMSE: 0.026RMSE: 0.048 LPIPS: 0.061
Ours+
RMSE: 0.011RMSE: 0.018RMSE: 0.025RMSE: 0.044 LPIPS: 0.038

10

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.043RMSE: 0.070RMSE: 0.024RMSE: 0.082 LPIPS: 0.068
Ours
RMSE: 0.048RMSE: 0.064RMSE: 0.069RMSE: 0.067 LPIPS: 0.059
[Gao19]+
RMSE: 0.033RMSE: 0.037RMSE: 0.017RMSE: 0.054 LPIPS: 0.029
Ours+
RMSE: 0.051RMSE: 0.035RMSE: 0.062RMSE: 0.055 LPIPS: 0.035

11

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.033RMSE: 0.012RMSE: 0.100RMSE: 0.145 LPIPS: 0.046
Ours
RMSE: 0.015RMSE: 0.026RMSE: 0.045RMSE: 0.059 LPIPS: 0.122
[Gao19]+
RMSE: 0.030RMSE: 0.008RMSE: 0.019RMSE: 0.021 LPIPS: 0.024
Ours+
RMSE: 0.011RMSE: 0.010RMSE: 0.029RMSE: 0.044 LPIPS: 0.028

12

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.044RMSE: 0.083RMSE: 0.078RMSE: 0.108 LPIPS: 0.037
Ours
RMSE: 0.043RMSE: 0.103RMSE: 0.049RMSE: 0.024 LPIPS: 0.042
[Gao19]+
RMSE: 0.023RMSE: 0.030RMSE: 0.029RMSE: 0.030 LPIPS: 0.012
Ours+
RMSE: 0.018RMSE: 0.037RMSE: 0.025RMSE: 0.020 LPIPS: 0.006

13

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.016RMSE: 0.032RMSE: 0.033RMSE: 0.049 LPIPS: 0.258
Ours
RMSE: 0.002RMSE: 0.023RMSE: 0.021RMSE: 0.032 LPIPS: 0.133
[Gao19]+
RMSE: 0.009RMSE: 0.017RMSE: 0.035RMSE: 0.044 LPIPS: 0.124
Ours+
RMSE: 0.003RMSE: 0.016RMSE: 0.019RMSE: 0.043 LPIPS: 0.067

14

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.023RMSE: 0.014RMSE: 0.028RMSE: 0.047 LPIPS: 0.085
Ours
RMSE: 0.017RMSE: 0.007RMSE: 0.020RMSE: 0.020 LPIPS: 0.011
[Gao19]+
RMSE: 0.018RMSE: 0.007RMSE: 0.016RMSE: 0.030 LPIPS: 0.019
Ours+
RMSE: 0.032RMSE: 0.007RMSE: 0.023RMSE: 0.016 LPIPS: 0.011

15

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.013RMSE: 0.008RMSE: 0.041RMSE: 0.047 LPIPS: 0.051
Ours
RMSE: 0.022RMSE: 0.012RMSE: 0.038RMSE: 0.011 LPIPS: 0.036
[Gao19]+
RMSE: 0.015RMSE: 0.008RMSE: 0.065RMSE: 0.046 LPIPS: 0.037
Ours+
RMSE: 0.007RMSE: 0.007RMSE: 0.054RMSE: 0.026 LPIPS: 0.028

16

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.102RMSE: 0.045RMSE: 0.032RMSE: 0.102 LPIPS: 0.129
Ours
RMSE: 0.083RMSE: 0.041RMSE: 0.030RMSE: 0.037 LPIPS: 0.085
[Gao19]+
RMSE: 0.059RMSE: 0.027RMSE: 0.023RMSE: 0.049 LPIPS: 0.075
Ours+
RMSE: 0.084RMSE: 0.019RMSE: 0.019RMSE: 0.035 LPIPS: 0.054

17

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.031RMSE: 0.083RMSE: 0.059RMSE: 0.056 LPIPS: 0.145
Ours
RMSE: 0.031RMSE: 0.085RMSE: 0.047RMSE: 0.024 LPIPS: 0.155
[Gao19]+
RMSE: 0.015RMSE: 0.019RMSE: 0.054RMSE: 0.078 LPIPS: 0.050
Ours+
RMSE: 0.010RMSE: 0.023RMSE: 0.077RMSE: 0.078 LPIPS: 0.078

18

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.070RMSE: 0.064RMSE: 0.287RMSE: 0.255 LPIPS: 0.141
Ours
RMSE: 0.055RMSE: 0.075RMSE: 0.340RMSE: 0.148 LPIPS: 0.062
[Gao19]+
RMSE: 0.058RMSE: 0.056RMSE: 0.120RMSE: 0.199 LPIPS: 0.126
Ours+
RMSE: 0.045RMSE: 0.059RMSE: 0.131RMSE: 0.214 LPIPS: 0.048

19

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.062RMSE: 0.060RMSE: 0.320RMSE: 0.257 LPIPS: 0.146
Ours
RMSE: 0.045RMSE: 0.070RMSE: 0.281RMSE: 0.094 LPIPS: 0.057
[Gao19]+
RMSE: 0.054RMSE: 0.059RMSE: 0.129RMSE: 0.172 LPIPS: 0.140
Ours+
RMSE: 0.042RMSE: 0.062RMSE: 0.133RMSE: 0.184 LPIPS: 0.053

20

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.011RMSE: 0.020RMSE: 0.027RMSE: 0.072 LPIPS: 0.204
Ours
RMSE: 0.002RMSE: 0.013RMSE: 0.020RMSE: 0.033 LPIPS: 0.076
[Gao19]+
RMSE: 0.009RMSE: 0.014RMSE: 0.021RMSE: 0.038 LPIPS: 0.152
Ours+
RMSE: 0.003RMSE: 0.007RMSE: 0.020RMSE: 0.030 LPIPS: 0.021

21

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.019RMSE: 0.015RMSE: 0.075RMSE: 0.059 LPIPS: 0.101
Ours
RMSE: 0.009RMSE: 0.013RMSE: 0.076RMSE: 0.020 LPIPS: 0.051
[Gao19]+
RMSE: 0.013RMSE: 0.013RMSE: 0.082RMSE: 0.088 LPIPS: 0.063
Ours+
RMSE: 0.008RMSE: 0.012RMSE: 0.062RMSE: 0.044 LPIPS: 0.049

22

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.030RMSE: 0.036RMSE: 0.029RMSE: 0.061 LPIPS: 0.084
Ours
RMSE: 0.004RMSE: 0.021RMSE: 0.018RMSE: 0.014 LPIPS: 0.015
[Gao19]+
RMSE: 0.025RMSE: 0.030RMSE: 0.020RMSE: 0.041 LPIPS: 0.048
Ours+
RMSE: 0.001RMSE: 0.011RMSE: 0.013RMSE: 0.012 LPIPS: 0.005

23

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.036RMSE: 0.011RMSE: 0.067RMSE: 0.179 LPIPS: 0.097
Ours
RMSE: 0.025RMSE: 0.012RMSE: 0.098RMSE: 0.169 LPIPS: 0.064
[Gao19]+
RMSE: 0.036RMSE: 0.007RMSE: 0.040RMSE: 0.085 LPIPS: 0.077
Ours+
RMSE: 0.024RMSE: 0.011RMSE: 0.054RMSE: 0.174 LPIPS: 0.069

24

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.023RMSE: 0.040RMSE: 0.026RMSE: 0.040 LPIPS: 0.070
Ours
RMSE: 0.022RMSE: 0.029RMSE: 0.032RMSE: 0.011 LPIPS: 0.035
[Gao19]+
RMSE: 0.020RMSE: 0.028RMSE: 0.022RMSE: 0.027 LPIPS: 0.039
Ours+
RMSE: 0.015RMSE: 0.019RMSE: 0.025RMSE: 0.010 LPIPS: 0.022

25

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.035RMSE: 0.057RMSE: 0.069RMSE: 0.093 LPIPS: 0.265
Ours
RMSE: 0.014RMSE: 0.031RMSE: 0.019RMSE: 0.022 LPIPS: 0.020
[Gao19]+
RMSE: 0.030RMSE: 0.046RMSE: 0.024RMSE: 0.073 LPIPS: 0.201
Ours+
RMSE: 0.009RMSE: 0.022RMSE: 0.021RMSE: 0.017 LPIPS: 0.014

26

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.077RMSE: 0.105RMSE: 0.047RMSE: 0.116 LPIPS: 0.097
Ours
RMSE: 0.027RMSE: 0.049RMSE: 0.015RMSE: 0.017 LPIPS: 0.013
[Gao19]+
RMSE: 0.046RMSE: 0.053RMSE: 0.017RMSE: 0.057 LPIPS: 0.056
Ours+
RMSE: 0.025RMSE: 0.034RMSE: 0.012RMSE: 0.018 LPIPS: 0.006

27

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.023RMSE: 0.018RMSE: 0.082RMSE: 0.293 LPIPS: 0.065
Ours
RMSE: 0.005RMSE: 0.015RMSE: 0.043RMSE: 0.178 LPIPS: 0.050
[Gao19]+
RMSE: 0.015RMSE: 0.011RMSE: 0.028RMSE: 0.131 LPIPS: 0.040
Ours+
RMSE: 0.003RMSE: 0.006RMSE: 0.037RMSE: 0.168 LPIPS: 0.017

28

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.056RMSE: 0.056RMSE: 0.433RMSE: 0.180 LPIPS: 0.257
Ours
RMSE: 0.023RMSE: 0.066RMSE: 0.040RMSE: 0.016 LPIPS: 0.092
[Gao19]+
RMSE: 0.048RMSE: 0.033RMSE: 0.077RMSE: 0.081 LPIPS: 0.224
Ours+
RMSE: 0.020RMSE: 0.041RMSE: 0.070RMSE: 0.106 LPIPS: 0.068

29

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.077RMSE: 0.042RMSE: 0.365RMSE: 0.554 LPIPS: 0.477
Ours
RMSE: 0.048RMSE: 0.050RMSE: 0.063RMSE: 0.409 LPIPS: 0.325
[Gao19]+
RMSE: 0.060RMSE: 0.031RMSE: 0.047RMSE: 0.235 LPIPS: 0.258
Ours+
RMSE: 0.050RMSE: 0.035RMSE: 0.075RMSE: 0.380 LPIPS: 0.178

30

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.038RMSE: 0.070RMSE: 0.178RMSE: 0.118 LPIPS: 0.142
Ours
RMSE: 0.030RMSE: 0.047RMSE: 0.053RMSE: 0.038 LPIPS: 0.056
[Gao19]+
RMSE: 0.035RMSE: 0.057RMSE: 0.072RMSE: 0.082 LPIPS: 0.085
Ours+
RMSE: 0.036RMSE: 0.043RMSE: 0.051RMSE: 0.051 LPIPS: 0.047

31

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.083RMSE: 0.127RMSE: 0.083RMSE: 0.103 LPIPS: 0.134
Ours
RMSE: 0.049RMSE: 0.104RMSE: 0.059RMSE: 0.018 LPIPS: 0.036
[Gao19]+
RMSE: 0.044RMSE: 0.086RMSE: 0.033RMSE: 0.073 LPIPS: 0.082
Ours+
RMSE: 0.044RMSE: 0.078RMSE: 0.062RMSE: 0.021 LPIPS: 0.017

32

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.030RMSE: 0.058RMSE: 0.090RMSE: 0.052 LPIPS: 0.122
Ours
RMSE: 0.027RMSE: 0.049RMSE: 0.078RMSE: 0.015 LPIPS: 0.055
[Gao19]+
RMSE: 0.027RMSE: 0.047RMSE: 0.078RMSE: 0.043 LPIPS: 0.086
Ours+
RMSE: 0.019RMSE: 0.041RMSE: 0.061RMSE: 0.012 LPIPS: 0.039

33

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.052RMSE: 0.083RMSE: 0.063RMSE: 0.065 LPIPS: 0.126
Ours
RMSE: 0.044RMSE: 0.076RMSE: 0.065RMSE: 0.013 LPIPS: 0.054
[Gao19]+
RMSE: 0.042RMSE: 0.079RMSE: 0.078RMSE: 0.066 LPIPS: 0.127
Ours+
RMSE: 0.026RMSE: 0.052RMSE: 0.052RMSE: 0.009 LPIPS: 0.017

34

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.065RMSE: 0.059RMSE: 0.141RMSE: 0.068 LPIPS: 0.040
Ours
RMSE: 0.092RMSE: 0.096RMSE: 0.085RMSE: 0.013 LPIPS: 0.054
[Gao19]+
RMSE: 0.028RMSE: 0.018RMSE: 0.077RMSE: 0.039 LPIPS: 0.017
Ours+
RMSE: 0.025RMSE: 0.025RMSE: 0.057RMSE: 0.011 LPIPS: 0.012

35

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.045RMSE: 0.039RMSE: 0.182RMSE: 0.047 LPIPS: 0.066
Ours
RMSE: 0.021RMSE: 0.032RMSE: 0.049RMSE: 0.010 LPIPS: 0.025
[Gao19]+
RMSE: 0.025RMSE: 0.024RMSE: 0.069RMSE: 0.047 LPIPS: 0.042
Ours+
RMSE: 0.013RMSE: 0.016RMSE: 0.039RMSE: 0.012 LPIPS: 0.016

36

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.048RMSE: 0.081RMSE: 0.160RMSE: 0.100 LPIPS: 0.089
Ours
RMSE: 0.053RMSE: 0.089RMSE: 0.173RMSE: 0.038 LPIPS: 0.069
[Gao19]+
RMSE: 0.041RMSE: 0.045RMSE: 0.101RMSE: 0.064 LPIPS: 0.054
Ours+
RMSE: 0.029RMSE: 0.040RMSE: 0.091RMSE: 0.028 LPIPS: 0.044

37

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.141RMSE: 0.140RMSE: 0.081RMSE: 0.102 LPIPS: 0.092
Ours
RMSE: 0.098RMSE: 0.113RMSE: 0.107RMSE: 0.027 LPIPS: 0.017
[Gao19]+
RMSE: 0.122RMSE: 0.111RMSE: 0.093RMSE: 0.106 LPIPS: 0.079
Ours+
RMSE: 0.065RMSE: 0.088RMSE: 0.115RMSE: 0.059 LPIPS: 0.010

38

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.033RMSE: 0.035RMSE: 0.214RMSE: 0.121 LPIPS: 0.224
Ours
RMSE: 0.008RMSE: 0.029RMSE: 0.029RMSE: 0.022 LPIPS: 0.053
[Gao19]+
RMSE: 0.012RMSE: 0.025RMSE: 0.033RMSE: 0.048 LPIPS: 0.082
Ours+
RMSE: 0.007RMSE: 0.020RMSE: 0.030RMSE: 0.008 LPIPS: 0.030

39

SVBRDF mapsPhotos/RenderingsPhotos/Renderings (novel)
Groundturth
[Deschaintre19]
[Gao19]
RMSE: 0.027RMSE: 0.024RMSE: 0.052RMSE: 0.081 LPIPS: 0.081
Ours
RMSE: 0.009RMSE: 0.012RMSE: 0.034RMSE: 0.014 LPIPS: 0.018
[Gao19]+
RMSE: 0.024RMSE: 0.015RMSE: 0.099RMSE: 0.068 LPIPS: 0.064
Ours+
RMSE: 0.009RMSE: 0.010RMSE: 0.027RMSE: 0.017 LPIPS: 0.012