AI Paper Atlas

Research map / Detection and segmentation / Weakly supervised detection and segmentation

Semi-Supervised and Long-Tailed Object Detection with CascadeMatch

Yuhang Zang, Kaiyang Zhou, Chen Huang, Chen Change Loy

IJCV 2023 · Detection and segmentation › Weakly supervised detection and segmentation · approach: Pseudo-labeling for semi-supervised detection · 15 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 cascade pseudo-labeling (CPL) (ablation)Supervised Cascade R-CNN · LVIS v1.0 valAP Fix: 26.3 -> 30.1Improved
Add adaptive pseudo-label mining (APM) alone (ablation)Supervised Cascade R-CNN · LVIS v1.0 valAP Fix rare: 19.7 -> 22.5Improved
Add APM on top of CPL (ablation)CascadeMatch w/ CPL · LVIS v1.0 valAP Fix: 30.1 -> 30.5Improved
Add CPL on top of APM (ablation)CascadeMatch w/ APM · LVIS v1.0 valAP Fix: 28.9 -> 30.5Improved
Use progressive epsilon 1.0/1.5/2.0 instead of all 0.0 (ablation)CascadeMatch · LVIS v1.0 valAP Fix: 29.8 -> 30.5Improved
Increase number of heads K from 1 to 3 (ablation)CascadeMatch · LVIS v1.0 valAP Fix: 26.4 -> 30.5Improved
Increase number of heads K from 3 to 5 (ablation)CascadeMatch · LVIS v1.0 valAP Fix rare: 23.1 -> 21.2Hurt
Increase unsupervised loss weight from 1.0 to 2.0 (ablation)CascadeMatch · LVIS v1.0 valAP Fix: 30.5 -> 29.4Hurt
Add CascadeMatch to Cascade R-CNN R-50 vs supervised baselineSupervised Cascade R-CNN · LVIS v1.0 valAP Fix: 26.3 -> 30.5Improved
Replace Unbiased Teacher with CascadeMatch (Sparse R-CNN PVT)Unbiased Teacher · LVIS v1.0 valAP Fix: 33.5 -> 35.2Improved
Add CascadeMatch to COCO-LT trainingSupervised Cascade R-CNN · COCO-LT valAP: 25.4 -> 27.8Improved
Add CascadeMatch under 40% missing labelsCascade R-CNN + Federated Loss · LVIS 40% annotations erasedAP: 22.5 -> 24.2Improved

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