Guiding a pretrained diffusion model toward per-class K-means centroids during sampling improves distilled dataset accuracy on ImageNet subsets by up to 4.4% while avoiding diffusion fine-tuning.
Method Labels IPC10 IPC20 IPC50 IPC 100 MGD3 (Ours) Hard-Label 23.6±0.4 32.6±0.4 51.8±0.2 57.8±0.2 Soft-label 34.0±1.0 50.2±0.7 69.2±0.4 75.8±0.3 I
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
-
MGD$^3$: Mode-Guided Dataset Distillation using Diffusion Models
Guiding a pretrained diffusion model toward per-class K-means centroids during sampling improves distilled dataset accuracy on ImageNet subsets by up to 4.4% while avoiding diffusion fine-tuning.