Search papers, labs, and topics across Lattice.
This paper introduces a saliency-driven framework for dataset distillation that enhances the representativeness and generalization of synthesized datasets by focusing on class-discriminative latent prototypes. By employing ensemble Grad-CAM saliency to identify high-confidence regions and refining these prototypes to ensure class consistency, the method addresses the common pitfalls of diffusion-based approaches that struggle with background noise and structural coherence. Experimental results across various benchmarks show significant performance improvements compared to existing methods, highlighting the effectiveness of the proposed technique in visually complex domains.
Saliency-driven prototype alignment can dramatically enhance dataset distillation, yielding better generalization and performance in complex visual tasks.
Dataset distillation aims to synthesize compact datasets that can approximate the performance of full-data training while significantly reducing computational and storage costs. However, diffusion-based distillation methods often struggle to preserve structural coherence and generalization, especially in visually complex domains. This issue often stems from latent prototypes that are weakly aligned with class-discriminative regions and contaminated by irrelevant background, thereby degrading generation quality and generalization. To address this limitation, we propose a saliency-driven distillation framework that constructs class-discriminative latent prototypes to enhance representativeness and generalization. The framework proceeds in two stages: (1) ensemble Grad-CAM saliency is used to construct prototypes emphasizing high-confidence regions, and (2) hard prototype refinement is then applied to construct challenging yet class-consistent prototypes, thereby enhancing discriminability and diversity. Importantly, the diffusion backbones (e.g., LDM and DiT) remain frozen; only lightweight classifiers used for saliency extraction are trained. Extensive experiments across multiple benchmarks demonstrate consistent performance improvements over strong baselines. Code will be released.
