DyWeight: Dynamic Gradient Weighting for Few-Step Diffusion Sampling

Dimensions overall score 9.0

GitHub Code Pulse

No public code linked for this paper yet.

• DyWeight introduces a lightweight, learning-based multi-step solver for diffusion sampling that uses a streamlined implicit coupling paradigm.method 95%
• DyWeight learns unconstrained time-varying parameters that adaptively aggregate historical gradients while intrinsically scaling the effective step size.method 95%
• DyWeight's implicit time calibration accurately aligns the solver's numerical trajectory with the model's internal denoising dynamics under large integration steps.technical 90%
• DyWeight achieves superior visual fidelity and stability with significantly fewer function evaluations compared to existing methods.result 90%
• DyWeight establishes a new state-of-the-art among efficient diffusion solvers.result 85%
• Existing multi-step ODE solvers for diffusion models rely on handcrafted coefficients that fail to adapt to the non-stationary dynamics of diffusion sampling.limitation 90%
• DyWeight avoids complex decoupled parameterizations and optimizations that other methods require.technical 85%
• DyWeight has been extensively tested on multiple datasets including CIFAR-10, FFHQ, AFHQv2, ImageNet64, LSUN-Bedroom, Stable Diffusion and FLUX.1-dev.result 95%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card