From 426803f15bb1db7d7ac74c124f9d2c5a46eb6263 Mon Sep 17 00:00:00 2001 From: wangshuai6 Date: Wed, 9 Apr 2025 11:41:22 +0800 Subject: [PATCH] update README --- README.md | 3 +++ 1 file changed, 3 insertions(+) diff --git a/README.md b/README.md index 0942ec0..fa84bbd 100644 --- a/README.md +++ b/README.md @@ -2,6 +2,9 @@ ## Introduction We decouple diffusion transformer into encoder-decoder design, and surpresingly that a **more substantial encoder yields performance improvements as model size increases**. ![](./figs/main.png) +* 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). +* As a byproduct, our DDT can reuse encoder among adjacent steps to accelerate inference. ## Visualizations ![](./figs/teaser.png) ## Usgae