AI Paper Atlas

Research map / Video / Video segmentation and tracking

Perceptual Video Coding for Machines via Satisfied Machine Ratio Modeling

Qi Zhang, Shanshe Wang, Xinfeng Zhang, Chuanmin Jia, Zhao Wang, Siwei Ma, Wen Gao

TPAMI 2024 · Video › Video segmentation and tracking · approach: Learned video compression · 14 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 the SMR-difference auxiliary task (model Q over G) (ablation)SMR prediction model · COCO-HEVC classification SMR-top1BD-rate (%): -29.2 -> -30.4Improved
add the SMR-difference auxiliary task (ablation)SMR prediction model · COCO-HEVC classification SMR-top5BD-rate (%): -43.4 -> -44.6Improved
add the SMR-difference auxiliary task (ablation)SMR prediction model · COCO-HEVC classification SMR-top1|dQP|: 1.40 -> 1.22Improved
add the SMR-difference auxiliary task (ablation)SMR prediction model · COCO-HEVC detection (T_IOU 0.5, T_S 0.5)BD-rate (%): -10.3 -> -12.3Improved
add the SMR-difference auxiliary task on an unseen codec (ablation)SMR prediction model · COCO-VVC detection (T_IOU 0.5, T_S 0.5)BD-rate (%): -4.8 -> -4.9No clear effect
add the SMR-difference auxiliary task on an unseen codec (ablation)SMR prediction model · COCO-VVC detection (T_IOU 0.6, T_S 0.6)BD-rate (%): -3.2 -> -0.2Hurt
use ground-truth SMR to choose QP per image instead of constant QPHEVC · COCO classification SMR-top1BD-rate (%): -39.2Improved
use ground-truth SMR to choose QPHEVC · COCO detectionBD-rate (%): -23.8, -23.2, -23.2 for three threshold pairsImproved
use predicted SMR on unseen machines (v2 library)SMR model G · COCO-HEVC classification SMR-top1BD-rate (%): -29.2 -> -30.2No clear effect
apply SMR models to unseen codec VVC instead of HEVCSMR model Q · COCO classification SMR-top1BD-rate (%): -30.4 -> -21.3Hurt
use more machines to compute SMR (Nm 10 -> 70) against full v2 librarySMR ground truth · COCO classification (v2 library)MAE of SMR-top1: 0.051 -> 0.004Improved
predict SMR-top5 instead of SMR-top1SMR model G · COCO-HEVC classification|dSMR|: 0.0463 -> 0.0217Improved

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