AI Paper Atlas

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

SILT: Shadow-Aware Iterative Label Tuning for Learning to Detect Shadows from Noisy Labels

Han Yang, Tianyu Wang, Xiaowei Hu, Chi-Wing Fu

ICCV 2023 · Label-efficient and robust learning › Noisy labels · approach: Noisy cross-modal hashing and multi-label · 20 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 global-local fusion to label tuning (ablation)SILT self-training · SBU (relabeled test)BER: 5.82 -> 4.67Improved
add shadow-aware filtering to self-training (ablation)ST · SBU (relabeled test)BER: 5.82 -> 5.44Improved
add shadow counterfeiting augmentation (ablation)ST+LT+SDA · SBU (relabeled test)BER: 4.46 -> 4.37Improved
naive self-training vs plain training on noisy labels (ablation)U-Net PVT v2-B5 · SBU (relabeled test)BER: 6.62 -> 5.82Improved
add 89 non-shadow dark-object images (ablation)SILT · SBUBER: 4.37 -> 4.19Improved
add 89 non-shadow dark-object images (ablation)SILT · UCF and ISTDBER: UCF 7.23 -> 7.60; ISTD 1.16 -> 1.18Hurt
train existing SOTA detectors on SILT-refined labelsBDRAR, DSC, FSD, FDRnet, SDCM · SBU (relabeled test)BER: SDCM 5.71 -> 4.87; FSD 10.87 -> 6.13Improved
use more self-training rounds than sevenSILT · SBUmask qualityHurt
set R_corr to 0.95 rather than 0.91 or 0.99 (ablation)SILT · SBUBER: 4.59 / 4.52 -> 4.19Improved
train on small well-labeled subset vs original noisy SBUSDCM · SBU splitBER: 5.30 -> 4.79Improved

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