ReSWD
Project Page | Paper | GitHub
ReSTIR‘d, not shaken. Combining Reservoir Sampling and Sliced Wasserstein Distance for Variance Reduction.
ReSWD is a method for distribution matching with reduced variance.
SW Guidance with stable-diffusion-3.5-large-turbo
Generates images using SW Guidance with a reference image and text prompt.
Input
Prompt + Reference Image Examples
| Prompt | Reference Image |
|---|
Pages:
Configuration
Adjust the parameters for the SW guidance process.
1 100
0 20
0 1
-6 0
0 1
0 32
16 1024
Control Variates
Select which control variates to use for optimization
Distance metric
Select which distance metric to use
0 64
1 16
Sampling Mode
Select which sampling mode to use for projections
Color Matching
Matches the color of source images to target images using ASC CDL parameters.
Source Images
Examples
Examples
Configuration
Adjust the parameters for the color matching process.
0.0001 0.1
64 1024
50 500
Control Variates
Select which control variates to use for optimization
0 64
16 1024
Sampling Mode
Select which sampling mode to use for projections