研究人员开发了UniGP，一个通过联合训练扩散Transformer模型来统一可控图像生成和密集预测任务的框架。该方法基于MMDiT，旨在捕捉图像-几何对的联合分布，而无需复杂的特定任务设计。实验表明，UniGP的表现与专用方法相当，并提供互补的优势，增强了生成中的感知细节和结构对齐。另外，AccelAes被提出作为一种无需训练的方法，通过将计算重新分配给具有更高美学相关性的区域来加速扩散Transformer以增强美学效果的图像生成。 AI
影响 扩散模型中的这些进展可能导致更高效、更多功能的图像创建和分析AI系统。
排序理由 该集群包含两篇研究论文，详细介绍了用于图像生成和感知任务的新AI模型和框架。
AI 生成摘要 · Google Gemini · 来自 4 个来源。 我们如何撰写摘要 →
arXiv:2607.00382v1 Announce Type: new Abstract: We propose the first compression approach for image-to-shape Diffusion Transformers (DiTs) that substantially reduces model size while preserving geometric fidelity. Despite remarkable progress in 3D shape generation, large DiT-base…
arXiv:2606.30332v1 Announce Type: new Abstract: Recent advances in diffusion models have shown impressive performance in controllable image generation and dense prediction tasks. However, existing approaches typically treat diffusion-based controllable generation and dense predic…
arXiv:2603.12575v2 Announce Type: replace Abstract: Diffusion Transformers (DiTs) are a dominant backbone for high-fidelity text-to-image generation due to strong scalability and alignment at high resolutions. However, quadratic self-attention over dense spatial tokens leads to h…
Recent advances in diffusion models have shown impressive performance in controllable image generation and dense prediction tasks. However, existing approaches typically treat diffusion-based controllable generation and dense prediction as separate tasks, overlooking the potentia…