Skip-DiT: Towards Stabilized and Efficient Diffusion Transformers through Long-Skip-Connections with Spectral Constraints
Published in International Conference on Computer Vision (ICCV), 2025
Skip-DiT introduces spectrally constrained long skip connections for more stable and efficient diffusion transformers, accelerating training by 4.4x and inference by up to 2x.
Recommended citation: Guanjie Chen, Xinyu Zhao, Yucheng Zhou, Xiaoye Qu, Tianlong Chen, and Yu Cheng. Skip-DiT: Towards Stabilized and Efficient Diffusion Transformers through Long-Skip-Connections with Spectral Constraints. ICCV, 2025.
Download Paper