Update README.md

README.md

@@ -17,21 +17,28 @@ datasets:
 pipeline_tag: unconditional-image-generation
 ---
 
-# Transformer AutoRegressive Flow Model
+**Transformer AutoRegressive Flow Model**
+
 The TarFlow is proposed by [Zhai et al., 2024], which introduces stacks of autoregressive Transformer blocks (similar to MAF) into the building of affine coupling layers to do Non-Volume Preserving, combined with guidance and denoising }, finally achieves state-of-the-art results across multiple benchmarks.
 
-Let $z$ denotes the noise direction and $x$ denotes the image direction, both with size $(B,T,C)$, where B,T,C represent batch size, patchified sequence length, and feature dimension, respectively. For TarFlow model, an autoregressive block can be written as:
+It's sampling process is extremely slow, and we want to accelerate it in []. In experiments, we find that the model parameters are not available in [Zhai et al., 2024], so we retrain TarFlow models and upload them.
+
+As metioned in [Zhai et al., 2024], a TarFlow model can be denoted as P-Ch-T-K-pε, with 
+patch size (P), model channel size (Ch), number of autoregressive flow blocks (T), the number of attention layers in each flow (K), the best input noise variance pε that yields the best sampling quality for generation tasks.
+
+We trained five models:
+
+ - AFHQ (256x256) conditional: afhq_model_8_768_8_8_0.07.pth
+
+ - ImageNet (128x128) conditional: imagenet_model_4_1024_8_8_0.15.pth
 
-\begin{equation}
-\begin{aligned}
-\text{Forward:\quad}z_t &= \exp(-s(x_{<t}))(x_t-u(x_{<t})),\\
-\text{Inverse:\quad}x_t &= \exp(s(x_{<t})) z_t +u(x_{<t}).
-\end{aligned}
-\end{equation}
+ - ImageNet (64x64) unconditional: imagenet64_model_2_768_8_8_0.05.pth
 
+ - ImageNet (64x64) conditional: imagenet_model_2_768_8_8_0.05.pth
 
-It's sampling process is extremely slow, and we want to accelerate it in []. In experiments, we found that the 
+ - ImageNet (64x64) conditioanl: imagenet_model_4_1024_8_8_0.05.pth
 
 
 
-[1] Zhai S, Zhang R, Nakkiran P, et al. Normalizing flows are capable generative models[J]. arXiv preprint arXiv:2412.06329, 2024.
+[1] Zhai S, Zhang R, Nakkiran P, et al. Normalizing flows are capable generative models[J]. arXiv preprint arXiv:2412.06329, 2024.
+[]