AI Paper Atlas

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

Solving Inefficiency of Self-Supervised Representation Learning

Guangrun Wang, Keze Wang, Guangcong Wang, Philip H.S. Torr, Liang Lin

ICCV 2021 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Contrastive versus non-contrastive theory · 69 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
use rank-5 negative instead of hardest negative (ablation)truncated triplet loss · ImageNet linear eval (20-epoch ablation)top-1 accuracy: 28.9 -> 29.5Improved
use rank-52 negative instead of hardest negative (ablation)truncated triplet loss · ImageNet linear eval (20-epoch ablation)top-1 accuracy: 28.9 -> 30.0Improved
use rank-52 instead of hardest triplet (k=1) (ablation)truncated triplet loss · over-clustering frequency at epoch 180Pr(Omega|B): 0.9948 -> 0.0233Improved
use k=5 instead of k=1 (ablation)truncated triplet loss · over-clustering frequency at epoch 180Pr(Omega|A): 0.9656 -> 0.2105Improved
use margin -100 instead of -0.3 (ablation)truncated triplet loss · ImageNet linear eval (20-epoch ablation)top-1 accuracy: 28.3 -> 30.0Improved
use margin -1.2 instead of -0.3 (ablation)truncated triplet loss · ImageNet linear eval (20-epoch ablation)top-1 accuracy: 28.3 -> 29.8Improved
use smoothed rank-k instead of rank-k triplet (ablation)truncated triplet loss · ImageNet linear eval (200 epochs)top-1 accuracy: 73.6 -> 73.8Improved
add multi-crop augmentation (ablation)truncated triplet loss · ImageNet linear eval (200 epochs)top-1 accuracy: 73.8 -> 74.1Improved
remove multi-crop from SwAVSwAV · ImageNet linear eval (200 epochs)top-1 accuracy: 72.7 -> 69.1Hurt
train truncated triplet longertruncated triplet loss · ImageNet linear evaltop-1 accuracy: 73.6 -> 75.9Improved
train supervised longersupervised ResNet-50 · ImageNettop-1 accuracy: 76.3 -> 78.4Improved
truncated triplet pretraining versus supervisedsupervised ImageNet pretraining · COCO 2017 Mask R-CNNAP box: 40.0 -> 41.7Improved

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