AI Paper Atlas

Research map / Label-efficient and robust learning / Noisy labels

NICEST: Noisy Label Correction and Training for Robust Scene Graph Generation

Lin Li, Jun Xiao, Hanrong Shi, Hanwang Zhang, Yi Yang, Wei Liu, Long Chen

TPAMI 2024 · Label-efficient and robust learning › Noisy labels · approach: Contrastive learning on noisy data · 43 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 Neg-NSD (mine missing negative samples) (ablation)Motifs · VG PredClsMean: 41.8 -> 43.8Improved
add Pos-NSD alone (ablation)Motifs · VG PredClsMean: 41.8 -> 39.8Hurt
add Neg-NSD and Pos-NSD without NSC (ablation)Motifs · VG PredClsMean: 41.8 -> 40.5Hurt
add NSC on top of Pos-NSD (ablation)Motifs · VG PredClsMean: 39.8 -> 42.4Improved
add Neg-NSD to Pos-NSD + NSC (ablation)Motifs · VG PredClsMean: 42.4 -> 44.3Improved
use full NICE vs baselineMotifs · VG PredClsmR@100: 17.8 -> 32.1Improved
add NIST multi-teacher distillation on NICEMotifs+NICE · VG PredClsMean: 44.3 -> 45.3Improved
add NIST on NICEMotifs+NICE · VG PredClsR@100: 58.6 -> 61.0Improved
add NIST on NICEMotifs+NICE · VG PredClsmR@100: 32.1 -> 31.6Hurt
add frequency prior (ablation)Motifs · VG-OOD PredClsMean: 34.9 -> 36.5 without priorHurt
add frequency prior (ablation)Motifs · VG PredClsMean: 40.9 -> 41.8Improved
NICE/NICEST vs label correction and label smoothingMotifs · VG PredClsMean: 3.9% ~ 4.7% absolute gainsImproved

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