AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

OAPT: Offset-Aware Partition Transformer for Double JPEG Artifacts Removal

Qiao Mo, Yukang Ding, Jinhua Hao, Qiang Zhu, Ming Sun, Chao Zhou, Feiyu Chen, Shuyuan Zhu

ECCV 2024 · Architectures and efficiency › Attention and transformers · approach: Efficient image transformers · 5 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 pattern-clustering attention with offset predictor to SwinIR backboneSwinIR · Classic5/LIVE1/BSDS500 grayscale double JPEGPSNR (dB): about +0.06 dB average over SwinIR; 1.5M extra parametersImproved
use ground-truth offsets instead of predicted offsets (ablation)OAPT · LIVE1 (70,40,5,1)PSNR (dB): 29.48 -> 29.53Improved
plug pattern clustering into OCAB of HAT-S with ground-truth offsetsHAT-S · double JPEG restorationPSNR (dB): +0.03 dB averageImproved
parallel hybrid attention instead of serial hybrid attention (ablation)OAPT HPAB · LIVE1 (30,50,4,4)PSNR (dB)Hurt
use uniform sparse attention only instead of serial hybrid (dense plus pattern-clustered) (ablation)OAPT HPAB · LIVE1 (30,50,4,4)PSNR (dB)Hurt
remove pattern clustering (uniform dense attention equals RSTB) (ablation)OAPT HPAB · LIVE1 (30,50,4,4)PSNR (dB)Hurt
offset prediction when QF1 >= QF2OAPT offset predictor · LIVE1 (90,10,2,0)offset accuracy: accuracy 97% at (10,90,0,2) vs 0% at (90,10,2,0)Hurt
train on double compression only then test on single compression (QF 10-50)OAPT · LIVE1 single compressionPSNR (dB): 29.75 -> 29.77 (QF=10 versus SwinIR)Improved
use ART-style uniform sparse attention instead of OAPTART · Classic5 (50,30,4,4)PSNR (dB): OAPT 32.97 vs ART 33.02Mixed

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