pixelddt-t2i update

This commit is contained in:
wangshuai6
2025-06-05 09:56:03 +08:00
parent 26c8e54edb
commit 910b764275

View File

@@ -15,7 +15,7 @@ We decouple diffusion transformer into encoder-decoder design, and surprisingly
* As a byproduct, our DDT can reuse encoder among adjacent steps to accelerate inference.
## Update 5/6/2025
* PixelDDT-XXL/16-R1024-T2I achieves **66.7** without prompt rewriting and **71.2** with prompt rewriting on GenEval benchmark
* Pixel Space Text-to-image models(PixelDDT-XXL/16) will be released soon.
* Pixel space Text-to-image models(PixelDDT-XXL/16) will be released soon.
## Visualizations
![](./figs/teaser.png)
## Checkpoints