AI Paper Atlas

Research map / Image generation / Image restoration

The Last Byte: Learning Just Enough for Machine-Oriented Image Compression

Wuyuan Xie, Zhenming Li, Ye Liu, Jian Jin, Yun Song, Miaohui Wang

AAAI 2026 · Image generation › Image restoration · approach: Learned image compression · 1 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 detector confidence map as input to the original image (ablation)MVRNet · MVRSet testE_total: 13.785 -> 7.735Improved
add original image to confidence-map-only input (ablation)MVRNet · MVRSet testE_total: 8.330 -> 7.735 (E_total); E_s 7.728 -> 8.250Mixed
add CAMM channel attention (ablation)MVRNet · MVRSet testE_total: 8.437 -> 7.844Improved
add CSFM fusion module (ablation)MVRNet · MVRSet testE_total: 8.437 -> 8.137Improved
combine CAMM and CSFM (ablation)MVRNet · MVRSet testE_total: 7.844 -> 7.735Improved
use single-pass no-reference MVRNet instead of OWJRDOWJRD · MVRSet testruntime: 7116.19s -> 0.17sImproved
use MVRNet instead of JRDVCMJRDVCM · MVRSet testE_total: 4.655 -> 4.140Improved
use MVRNet vs H.266/VVC at COCO detectionH.266/VVC · COCOmAP@.50 / BPP: 5.8720 BPP/0.7290 -> 0.5314 BPP/0.8640 mAPImproved
use MVRNet with DETR detector vs H.266/VVCH.266/VVC · COCO detection (DETR)mAP@.50: 0.7290 -> 0.7100 mAP@.50, BPP 5.8720 -> 0.5435No clear effect
use MVRNet for instance segmentation vs H.266/VVCH.266/VVC · COCO instance segmentationmAP@.50: 0.6640 -> 0.6341 mAP@.50, BPP 5.8643 -> 0.4481Mixed

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