The authors modified MiniT2I, a pixel-space multimodal diffusion Transformer, by dividing its blocks into three groups.
Reusing Transformer blocks improves image generation without enlarging the model
Looped-DiT repeatedly applies shared layers within each denoising step, using intermediate supervision and gated attention to make that added computation effective.
Chinese Tech
Yong Xien Chng · Tianyi Chen · Wenwen Tong · Haiwen Diao · Zhongang Cai · Lei Yang · +4 more
SenseTime Research · LeapLab, Tsinghua University · Nanyang Tehnological Univesity
Research Digest··2 min read
Chng et al.
Why this paper
From SenseTime Research and 2 others
In one line
Looped diffusion transformers with deep supervision and self-modulating attention outperform much larger models with less compute.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§