Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper identifies resolution-dependent parameter behaviors in diffusion models as the cause of performance degradation at out-of-training resolutions. To address this, they introduce CR-Diff, a method that selectively prunes adverse weights in the UNet architecture through block-wise pruning and then amplifies the pruned output. Experiments demonstrate that CR-Diff improves perceptual fidelity and semantic coherence across various diffusion backbones and unseen resolutions, while maintaining performance at default resolutions.
Diffusion models can generate higher quality images at arbitrary resolutions by selectively pruning weights that become "adverse" when spatial scales shift.
Diffusion models have demonstrated impressive image synthesis performance, yet many UNet-based models are trained at certain fixed resolutions. Their quality tends to degrade when generating images at out-of-training resolutions. We trace this issue to resolution-dependent parameter behaviors, where weights that function well at the default resolution can become adverse when spatial scales shift, weakening semantic alignment and causing structural instability in the UNet architecture. Based on this analysis, this paper introduces CR-Diff, a novel method that improves the cross-resolution visual consistency by pruning some parameters of the diffusion model. Specifically, CR-Diff has two stages. It first performs block-wise pruning to selectively eliminate adverse weights. Then, a pruned output amplification is conducted to further purify the pruned predictions. Empirically, extensive experiments suggest that CR-Diff can improve perceptual fidelity and semantic coherence across various diffusion backbones and unseen resolutions, while largely preserving the performance at default resolutions. Additionally, CR-Diff supports prompt-specific refinement, enabling quality enhancement on demand.