AI Paper Atlas

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

PreDet: Large-Scale Weakly Supervised Pre-Training for Detection

Vignesh Ramanathan, Rui Wang, Dhruv Mahajan

ICCV 2021 · Detection and segmentation › Weakly supervised detection and segmentation · approach: Weakly supervised detection · 13 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 detection-specific query-box lookup losses to IG-50M classification pre-trainingMask R-CNN ResNeXt-101-32x8d · MS-COCO 1xAPbox: 44.4 -> 47.1Improved
add classification loss Lcls to the detection losses (ablation)PreDet · MS-COCO 1xAPbox: 43.7 -> 47.1Improved
add CROI-head losses to the classification-only model (ablation)Lcls model · MS-COCO 1xAPbox: 44.6 -> 46.3Improved
add CROI regression loss (ablation)PreDet CROI-head · MS-COCO 1xAPbox: 45.9 -> 46.3Improved
add CRPN losses (ablation)PreDet · MS-COCO 1xAPbox: 46.3 -> 47.1Improved
pre-train on IG-50M instead of ImageNetPreDet · MS-COCO 1xAPbox: 45.8 -> 47.1Improved
PreDet-IG50M vs InfoMin on localizationMask R-CNN · MS-COCOAPbox75: 49.9 -> 51.6Improved
PreDet pre-training vs training from scratch at full scheduleMask R-CNN · MS-COCOAPbox: 45.8 -> 47.1Improved
use PreDet-IG50M instead of best baseline on RetinaNetRetinaNet · MS-COCOAPbox: 43.0 -> 45.1Improved
PreDet-IG50M instead of SEER on PASCAL VOCFaster R-CNN · PASCAL VOC07 testAPbox: 58.6 -> 62.5Improved
PreDet-IG50M instead of ImageNet pre-training with small target setMask R-CNN · MS-COCO 1k imagesAPbox: 11.3 -> 16.1Improved
larger backbone capacity (101-32x16d vs 50-32x4d) (ablation)PreDet vs ImageNet · MS-COCO 1xAPbox gain: 2.3% -> 3.6%Improved

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