AI Paper Atlas

Research map / Image generation / Super-resolution

Fine-structure Preserved Real-world Image Super-resolution via Transfer VAE Training

Qiaosi Yi, Shuai Li, Rongyuan Wu, Lingchen Sun, Yuhui Wu, Lei Zhang

ICCV 2025 · Image generation › Super-resolution · approach: Diffusion-based super-resolution · 12 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add Transfer VAE Training (align latent space with VAE-D8) instead of training VAE-D4 from scratch (ablation)VAE-D4 Real-ISR · RealSRPSNR: 25.50 -> 25.83Improved
add Transfer VAE Training instead of L1 latent alignment (ablation)VAE-D4 Real-ISR · RealSRCLIPIQA: 0.6734 -> 0.6924Improved
add TVT versus training VAE-D4 from scratch (ablation)VAE-D4 Real-ISR · RealSRMUSIQ: 67.85 -> 69.26Improved
use compact VAE-D4 (no middle block, 128/256/256 channels) instead of larger V1 (ablation)VAE-D4 · RealSRFLOPs: 3.02 T -> 1.97 T; PSNR 25.84 -> 25.83No clear effect
use compact VAE-D4 instead of V2 (ablation)VAE-D4 · RealSRSSIM: 0.7603 -> 0.7619No clear effect
replace VAE-D8 with VAE-D4 (full UNet) (ablation)one-step Real-ISR · RealSRSSIM: 0.7419 -> 0.7632Improved
replace VAE-D8 with VAE-D4 (full UNet) (ablation)one-step Real-ISR · RealSRQ-Align: 3.592 -> 3.797Improved
add skip connections to VAE-D8 (ablation)VAE-D8 Real-ISR · RealSRQ-Align: 3.592 -> 3.482Hurt
use 8-channel latent VAE-D8 instead of 4-channel (ablation)VAE-D8 Real-ISR · RealSRMUSIQ: 68.86 -> 67.79Hurt
use compute-efficient UNet instead of 25% channel-pruned UNet (ablation)VAE-D4 Real-ISR · RealSRSSIM: 0.7446 -> 0.7619Improved
use compute-efficient UNet instead of full UNet with VAE-D4 (ablation)VAE-D4 Real-ISR · RealSRFLOPs: 2.59 T -> 1.97 TImproved
replace VAE-D8 with adapted VAE-D4 for reconstructionVAE · Urban100 (VAE reconstruction)PSNR: 25.29 -> 32.35Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy