AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

Efficient Visual Pretraining With Contrastive Detection

Olivier J. Hénaff, Skanda Koppula, Jean-Baptiste Alayrac, Aaron van den Oord, Oriol Vinyals, João Carreira

ICCV 2021 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Invariance-based self-supervised learning · 145 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
replace global pooling with a single FH mask per image (ablation)SimCLR · COCO pretrain, COCO frozen-featureAP bbox: 31.6 -> 31.2Hurt
duplicate the global mask 16 times (ablation)SimCLR · COCO pretrain, COCO frozen-featureAP bbox: 31.6 -> 31.5No clear effect
sample 16 FH masks per image (dense mask sampling) (ablation)SimCLR with FH masks · COCO pretrain, COCO frozen-featureAP bbox: 31.2 -> 33.4Improved
replace SimCLR with full DetCon_SSimCLR · COCO pretrain, COCO frozen-featureAP bbox: 31.6 -> 33.4Improved
replace SimCLR with full DetCon_SSimCLR · COCO pretrain, COCO frozen-featureAP mask: 29.2 -> 30.6Improved
gather negatives from the entire batch (FH masks) (ablation)DetCon_S · COCO frozen-feature, FH masksAP bbox: 32.2 -> 33.6Improved
gather negatives from the entire batch (GT masks) (ablation)DetCon_S · COCO frozen-feature, GT masksAP bbox: 38.5 -> 37.0Hurt
make predictions across two augmented views (FH masks) (ablation)DetCon_S · COCO frozen-feature, FH masksAP bbox: 27.7 -> 32.2Improved
make predictions across two augmented views (GT masks) (ablation)DetCon_S · COCO frozen-feature, GT masksAP bbox: 38.8 -> 38.5Hurt
use DetCon_B instead of BYOL pretraining (300 epochs)BYOL · PASCAL segmentationmIoU: 74.7 -> 75.6Improved
use DetCon_B instead of BYOL pretraining (300 epochs)BYOL · Cityscapes segmentationmIoU: 73.4 -> 75.1Improved
use DetCon_B instead of BYOL (1x fine-tune)BYOL · COCO detectionAP bbox: 41.6 -> 42.7Improved

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