Images

Use mouse wheel to zoom in/out, click and drag to pan. Press keys [1], [2], ... to switch between individual images.

4 spp

16 spp

32 spp

64 spp

128 spp

Color

1-SSIM

RMSE

1: Input

2: NFOR

3: RAE

4: KPCN

5: Ours

6: Reference

Input

NFOR

RAE

KPCN

Ours

Reference

Charts

Relative performance, obtained by dividing the reconstruction error by the default path traced rendering error. For each method, we show, from left to right, the improvement at 4, 16, 32, 64 and 128 samples per pixel. For RMSE and 1-SSIM, lower values are better. For PSNR, higher values are better.

1-SSIM

PSNR

RMSE

Error metrics

1-SSIM
Setting	Input	NFOR	RAE	KPCN	Ours
4 spp	0.13394	0.03316	0.02070	0.01728	0.01640
16 spp	0.09631	0.01563	0.01111	0.00885	0.00868
32 spp	0.07706	0.01010	0.00848	0.00638	0.00464
64 spp	0.05799	0.00650	0.00643	0.00464	0.00381
128 spp	0.04139	0.00443	0.00487	0.00348	0.00343

PSNR
Setting	Input	NFOR	RAE	KPCN	Ours
4 spp	21.80008	33.78439	34.95014	34.08797	35.71009
16 spp	27.67337	38.45872	39.15719	39.08629	40.28707
32 spp	31.26659	40.71328	41.01803	41.49452	43.73760
64 spp	34.86356	43.00562	42.90684	43.76791	44.85177
128 spp	38.08044	44.92254	44.78014	45.84639	45.35287

RMSE
Setting	Input	NFOR	RAE	KPCN	Ours
4 spp	0.05376	0.00628	0.00414	0.00309	0.00503
16 spp	0.01762	0.00233	0.00147	0.00140	0.00204
32 spp	0.00801	0.00124	0.00097	0.00090	0.00092
64 spp	0.00356	0.00073	0.00068	0.00061	0.00072
128 spp	0.00171	0.00044	0.00046	0.00039	0.00060