Diffusion models can keep most of their image quality after swapping global self-attention for a distilled multi-scale convolutional block, with FLOPs claimed to fall by up to 6929 times at 16K resolution.
Simple baselines for image restoration
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Can We Achieve Efficient Diffusion without Self-Attention? Distilling Self-Attention into Convolutions
Diffusion models can keep most of their image quality after swapping global self-attention for a distilled multi-scale convolutional block, with FLOPs claimed to fall by up to 6929 times at 16K resolution.