网格蒸馏：基于结构化生成网格的组合式图像蒸馏
Grid Distillation: Compositional Image Distillation via Structured Generative Grids
把数据集蒸馏的合成样本组织成高分辨率组合网格、再降采样并用 Stable Diffusion Turbo 单步重建细节，这套「网格布局 + 扩散先验补全」思路可迁移到任何需要把大量图像压缩成少量可训练样本的场景（如快速架构搜索、隐私受限数据发布）。
原文依据 a single-step diffusion reconstruction (based on Stable Diffusion Turbo) restores fine-grained spatial details from diffusion priors
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论本文提出网格蒸馏（Grid Distillation），一种生成式数据集蒸馏框架，可将大规模数据集压缩为一组紧凑且信息量丰富的合成样本。该方法通过谱子模优化构建高分辨率组合网格，并注入来自 CLIP 表示的世界知识，以最大化语义覆盖度与多样性。随后，这些网格被降采样为低分辨率蒸馏图像，并针对多样性与表示效率进行优化。在训练阶段，基于 Stable Diffusion Turbo 的单步扩散重建利用扩散先验恢复细粒度空间细节，从而弥合紧凑表示与自然图像统计特性之间的差距。一种网格感知的裁剪策略通过使裁剪区域按概率与网格边界对齐来进一步增强判别性，同时保持与标准 224×224 推理输入的兼容性。在 ImageWoof、ImageNette、ImageIDC 和 ImageNet-1K 上的实验表明，在多种 IPC 设置下，该方法相较现有数据集蒸馏方法均取得了一致的提升。
英文原文摘要 1081 字符
We present Grid Distillation, a generative dataset distillation framework that compresses large-scale datasets into a compact set of informative synthetic samples. Our method constructs high-resolution compositional grids via spectral submodular optimization, which injects world knowledge from CLIP representations to maximize semantic coverage and diversity. These grids are then downsampled into low-resolution distilled images optimized for diversity and representational efficiency. During training, a single-step diffusion reconstruction (based on Stable Diffusion Turbo) restores fine-grained spatial details from diffusion priors, bridging the gap between compact representations and natural image statistics. A grid-aware cropping strategy further enhances discriminability by probabilistically aligning crops with grid boundaries, maintaining compatibility with standard 224 x 224 inference inputs. Experiments on ImageWoof, ImageNette, ImageIDC, and ImageNet-1K demonstrate consistent improvements over existing dataset distillation methods across multiple IPC settings.