Update README.md

README.md

@@ -48,15 +48,19 @@ pipe = QwenImage21Pipeline.from_pretrained("Qwen/Qwen-Image-2.1", vae=vae, torch
 
 # Mechanism
 
-Qwen-Image-2.1-VAE-Texture-Fix was created by finetuning the Qwen-Image-2.1 VAE decoder for ~5000 steps at learning rate 3e-5, with only the two highest-resolution decoder stages and output head unfrozen (7.5M trainable parameters), using the recipe developed for [TAESD](https://github.com/madebyollin/taesd). This recipe includes an adversarial loss term which strongly penalizes checkerboarding (and other obvious texture artifacts).
+Qwen-Image-2.1-VAE-Texture-Fix was created by finetuning the Qwen-Image-2.1 VAE decoder for ~5000 steps at learning rate 3e-5, with only the two highest-resolution decoder stages and output head unfrozen (7.5M trainable parameters), using the recipe developed for [TAESD](https://github.com/madebyollin/taesd).
 
-VAEs are usually trained with a mix of PSNR-focused (MSE/MAE), LPIPS, and adversarial loss terms. Whenever precise details can't be reconstructed, MSE/MAE loss encourages blurring, LPIPS loss encourages blurring+checkerboarding (and various other artifacts), and adversarial loss encourages generating sharp/plausible (but fake) detail without obvious artifacts.
+The TAESD recipe, like most image autoencoder training recipes, uses a mix of PSNR-focused (MSE/MAE), LPIPS, and adversarial (GAN) loss terms.
+Whenever precise details can't be reconstructed, MSE/MAE loss encourages blurring, LPIPS loss encourages blurring+checkerboarding (among other artifacts), and adversarial loss encourages generating sharp/plausible (but fake) detail without obvious artifacts.
+This figure from DC-AE (https://arxiv.org/abs/2410.10733) shows the importance of including adversarial (GAN) loss:
 
-I suspect the original Qwen-Image-2.1-VAE was trained without adversarial loss (or with a very weak/nonfunctioning adversarial loss).
+![Demo of the effects of adversarial loss, courtesy of the DC-AE paper](https://cdn-uploads.huggingface.co/production/uploads/630447d40547362a22a969a2/uj9vJSR34LyZEkQeL-VOU.png)
+
+I suspect the original Qwen-Image-2.1-VAE was trained without a working adversarial loss term.
 
 # Metrics
 
-Qwen-Image-2.1-VAE-Texture-Fix makes perceptual quality metrics (rFID) better and reconstruction accuracy metrics (LPIPS/PSNR) worse.
+Qwen-Image-2.1-VAE-Texture-Fix makes perceptual quality metrics (rFID) better and reconstruction accuracy metrics (LPIPS/PSNR) slightly worse.
 
 | Metric | Qwen-Image-2.1-VAE | Qwen-Image-2.1-VAE-Texture-Fix |
 | --- | --- | --- |