DiT-Distill：通过生成式课程知识实现开放集细粒度检索
DiT-Distill: Open-Set Fine-Grained Retrieval via Generative Curriculum Knowledge
把 DiT 多时间步的生成式知识通过课程对齐损失蒸馏进轻量骨干网络，实现免 DiT 推理，为视觉检索/表征学习提供了一条「用生成模型当教师、按扩散时间步组织课程」的可复用蒸馏范式。
原文依据 transfers this refined, hierarchical knowledge from multiple diffusion timesteps of the DiT into a lightweight backbone
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论开放集细粒度检索（OSFR）是一项具有挑战性的任务，模型必须泛化到未见过的子类别。现有方法往往在此失败，因为它们嵌入了来自闭集训练标签的类别特定语义。近来，扩散 Transformer（DiT）展现出潜力，其编码的是以属性为中心、与这些标签无关的生成式课程知识。然而，原始 DiT 并未针对细粒度视觉差异进行优化，且其庞大参数量使部署不可行。为解决这一问题，我们提出 DiT-Distill，一个先精炼再蒸馏该知识的框架。我们引入条件差异精炼策略对 DiT 进行微调，迫使其关注差异感知、以属性为中心的细节，而非整体上下文。随后，生成式课程蒸馏机制通过生成式注入模块与课程对齐损失，将这一精炼后的层次化知识从 DiT 的多个扩散时间步迁移到轻量骨干网络中。该过程得到一个高效的检索模型，可实现免 DiT 推理。大量实验表明，DiT-Distill 在开放集细粒度数据集上取得了最优性能。
英文原文摘要 1234 字符
Open-set fine-grained retrieval (OSFR) is a challenging task where models must generalize to unseen subcategories. Existing methods often fail this, as they embed category-specific semantics from closed-set training labels. Recently, diffusion transformers (DiT) have shown promise by encoding attribute-centric, generative curriculum knowledge that is agnostic to these labels. However, the vanilla DiT is not optimized for fine-grained visual discrepancies and its massive size makes deployment infeasible. To solve this, we propose DiT-Distill, a framework to first refine and then distill this knowledge. We introduce a conditional discrepancy refinement strategy to fine-tune the DiT, forcing it to focus on discrepancy-aware, attribute-centric details rather than holistic context. Subsequently, a generative curriculum distillation mechanism transfers this refined, hierarchical knowledge from multiple diffusion timesteps of the DiT into a lightweight backbone using a generative infusion module and a curriculum alignment loss. This process results in an efficient retrieval model that enables DiT-free inference. Extensive experiments show DiT-Distill achieves state-of-the-art performance on open-set fine-grained datasets.