AI Paper Atlas

Research map / Image generation / Image restoration

Analyzing the Training Dynamics of Image Restoration Transformers: A Revisit to Layer Normalization

MinKyu Lee, Sangeek Hyun, Woojin Jun, Hyunjun Kim, Jiwoo Chung, Jae-Pil Heo

ICLR 2026 · Image generation › Image restoration · approach: Priors for restoration networks · 2 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
replace LayerNorm with i-LNHAT_1 · x4 SR Urban100PSNR: 26.55 -> 27.17Improved
replace LayerNorm with i-LNHAT_1 · Rain100L derainingPSNR: 34.35 -> 36.20Improved
replace LayerNorm with i-LNHAT_1 · Urban100 denoising sigma=15PSNR: 35.489 -> 35.558Improved
replace LayerNorm with i-LNDRCT_1 · x2 SR Set5PSNR: 38.19 -> 38.23No clear effect
add spatial holisticness (ablation)i-LN · x4 SR Urban100 (HAT_2)PSNR: 26.8779 -> 27.5335Improved
add rescaling to spatial holistic norm (ablation)i-LN · x4 SR Urban100 (HAT_2)PSNR: 27.5335 -> 27.5849Improved
use InstanceNorm instead of i-LN (ablation)HAT_1 · x4 SR Urban100PSNR: 27.17 -> 27.02Hurt
use per-token variants (LayerScale, RMSNorm) instead of LN (ablation)HAT_1 · x4 SR Urban100PSNR: 26.55 -> 26.75 (LayerScale)Improved
use ReZero instead of LN (ablation)HAT_1 · x4 SR Urban100PSNR: 26.55 -> 26.41Hurt
remove all normalization (ablation)HAT_1 · x4 SRPSNRHurt
replace LN with i-LN under int4 weight quantizationHAT_2 · x4 SR Urban100PSNR: 25.0242 -> 26.8292Improved
replace LN with i-LN under fp16 inferenceHAT_2 · x4 SR Urban100PSNR: 7.4640 -> 27.5849Improved

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