Logic/DDT
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

pixelddt-t2i update

Browse Source
This commit is contained in:
wangshuai6
2025-06-05 09:55:43 +08:00
parent 99d92c94e7
commit 26c8e54edb
1 changed files with 3 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
README.md
View File

@@ -13,6 +13,9 @@ We decouple diffusion transformer into encoder-decoder design, and surprisingly
* We achieves **1.26 FID** on ImageNet256x256 Benchmark with DDT-XL/2(22en6de). * We achieves **1.26 FID** on ImageNet256x256 Benchmark with DDT-XL/2(22en6de).
* We achieves **1.28 FID** on ImageNet512x512 Benchmark with DDT-XL/2(22en6de). * We achieves **1.28 FID** on ImageNet512x512 Benchmark with DDT-XL/2(22en6de).
* As a byproduct, our DDT can reuse encoder among adjacent steps to accelerate inference. * 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.
## Visualizations ## Visualizations
![](./figs/teaser.png) ![](./figs/teaser.png)
## Checkpoints ## Checkpoints