ProcessMaker：基于扩散 Transformer 的自适应序列步骤通用过程可视化框架
ProcessMaker: A Generalized Process Visualization Framework with Adaptive Sequence Steps on Diffusion Transformers
把 DiT 的泛化能力用于多步过程序列生成，用自监督表征对齐替代多专家网络、用稀疏掩码做域区分、用滑动窗口动态决定生成步数，这套「单模型多域+自适应步数」思路可迁移到需要变长生成轨迹的视觉生成任务。
原文依据 a novel framework that harnesses the inherent generalization capabilities in Diffusion Transformers (DiTs) for procedural sequence generation
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论过程序列生成旨在通过多步过程生成中间图像，应用于工业设计、教学教程和书籍插图。然而，现有方法往往聚焦于特定领域，或为不同领域初始化多个专家网络，因而面临三个挑战：第一，对未见领域的泛化能力差；第二，多专家网络带来的参数冗余；第三，难以针对不同过程自适应地确定生成步数。为解决这些挑战，我们提出 ProcessMaker，一种新颖的框架，利用扩散 Transformer（DiT）固有的泛化能力进行过程序列生成。具体而言，我们引入三项关键创新：（1）自监督表征对齐，以挖掘对未见过程的泛化能力；（2）面向不同领域的稀疏掩码，无需额外的专家网络；（3）滑动窗口策略，根据过程复杂度动态适配生成步数。大量实验验证，ProcessMaker 在具备泛化能力和自适应步数的同时实现了过程序列生成，且仅使用最优方法 7.3% 的可训练参数。
英文原文摘要 1311 字符
Procedural sequence generation aims to create intermediate images through multi-step processes, which is applied in industrial design, educational tutorial, and book illustration. However, existing methods often focus on a specific domain or initialize several expert networks for different domains, which face three challenges. First, the poor generalization to unseen domains. Second, the parameter redundancy due to multiple expert networks.Third, the difficulty in adaptively determining the number of generation steps for different processes.To address these challenges, we propose ProcessMaker, a novel framework that harnesses the inherent generalization capabilities in Diffusion Transformers (DiTs) for procedural sequence generation. Concretely, we introduce three key innovations: (1) Self-supervised Representation Alignment to explore the generalized ability for unseen processes. (2) Sparse Masks for different domains without additional expert networks. (3) A sliding window strategy, which dynamically accommodates the generation steps based on the process complexity. Extensive experiments validate that our ProcessMaker achieves procedural sequence generation with generalization ability and adaptive steps, while using only 7.3% trainable parameters compared with the state-of-the-art method.