DiT-Air: Revisiting the Efficiency of Diffusion Model Architecture Design in Text to Image Generation
December 11, 2025 research area Computer Vision:/research/?domain=Computer%20Vision
In this work, we empirically study Diffusion Transformers (DiTs) for text-to-image generation, focusing on architectural choices, text-conditioning strategies, and training protocols. We evaluate a range of DiT-based architectures—including PixArt-style and MMDiT variants—and compare them with a standard DiT variant which directly processes concatenated text and noise inputs. Surprisingly, our findings reveal that the performance of standard…
On Inductive Biases That Enable Generalization of Diffusion Transformers
September 22, 2025 research area Computer Vision:/research/?domain=Computer%20Vision conference NeurIPS:/research/?event=NeurIPS
Recent work studying the generalization of diffusion models with UNet-based denoisers reveals inductive biases that can be expressed via geometry-adaptive harmonic bases. However, in practice, more recent denoising networks are often based on transformers, e.g., the diffusion transformer (DiT). This raises the question: do transformer-based denoising networks exhibit inductive biases that can also be expressed via geometry-adaptive harmonic…
Our research in machine learning breaks new ground every day.
情报判断
Aioga 编辑摘要
研究聚焦扩散 Transformer 图像生成中的离群 token:预训练 ViT 编码器与 DiT 去噪器都会出现该现象,中间层尤为明显。论文提出 Dual-Stage Registers,并报告其在 ImageNet 和文生图任务上减少离群伪影、提升生成质量。