AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Feature Coding in the Era of Large Models: Dataset, Test Conditions, and Benchmark

Changsheng Gao, Yifan Ma, Qiaoxi Chen, Yenan Xu, Dong Liu, Weisi Lin

ICCV 2025 · Label-efficient and robust learning › Recognition and long-tail · approach: Correspondence and causal representation learning · 16 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
use VTM codec instead of Hyperprior as the core feature codecfeature coding baseline · Image classificationAccuracy near 2 BPFP: 93.20 -> 99.80Improved
use VTM instead of Hyperprior for text featuresfeature coding baseline · Common sense reasoningDistortion-accuracy R^2: 0.4713 -> 0.7490Improved
reduce bitrate by raising QP in VTM baselineVTM baseline · Image classificationAccuracy: 99.80 -> 7.20Hurt
use feature MSE as a proxy for task accuracyboth baselines · Common sense reasoning (text features)Distortion-accuracy R^2: 0.4713Hurt
use image codecs on generated-image featuresboth baselines · Text-to-image synthesisDistortion-accuracy R^2: 0.9983 and 0.9996Improved
apply a learned codec trained on another task's featuresHyperprior baseline · Cross-task generalization (Seg, CSR, TTI)Accuracy / FIDHurt
run VTM encoder on large-model featuresVTM baseline · ComplexityEncoding timeHurt
replace Bits Per Pixel with Bits Per Feature Point as bitrate measuretest conditions · metric designbitrate fairnessImproved

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