AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Parametric Contrastive Learning

Jiequan Cui, Zhisheng Zhong, Shu Liu, Bei Yu, Jiaya Jia

ICCV 2021 · Label-efficient and robust learning › Recognition and long-tail · approach: Decoupled and calibrated long-tail training · 334 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 cross-entropy with supervised contrastive lossResNet-50 · ImageNet-LTtop-1 accuracy: 48.4 -> 22.0Hurt
add parametric class-wise learnable centers to supervised contrastive loss (ablation)SupCon, ResNet-50 · ImageNet-LTtop-1 accuracy: 22.0 -> 51.0Improved
random resized crop (0.08,1) instead of (0.2,1) (ablation)PaCo, ResNet-50 · ImageNet-LTtop-1 accuracy: 52.2 -> 55.0Improved
RandAugment on encoder and SimAugment on momentum encoder instead of SimAugment on both (ablation)PaCo, ResNet-50 · ImageNet-LTtop-1 accuracy: 55.0 -> 57.0Improved
RandAugment on encoder with SimAugment on momentum encoder instead of RandAugment on both (ablation)PaCo, ResNet-50 · ImageNet-LTtop-1 accuracy: 56.5 -> 57.0Improved
PaCo instead of Balanced SoftmaxResNet-50 · ImageNet-LT (RandAugment, 400 epochs)top-1 accuracy: 55.0 -> 57.0Improved
PaCo instead of CE plus re-weighted contrastive multi-task (ablation)ResNet-50 · ImageNet-LTtop-1 accuracy: 49.0 -> 51.0Improved
PaCo instead of multi-task Balanced Softmax plus re-weighting (loss weight 0.05) (ablation)ResNeXt-50 · ImageNet-LTtop-1 accuracy: 57.0 -> 58.2Improved
PaCo instead of supervised contrastive learningResNet-50 · CIFAR-100 (full)top-1 accuracy: 76.5 -> 79.1Improved
PaCo instead of multi-task CE plus contrastive (ablation)ResNet-50 · ImageNet (full, RandAugment)top-1 accuracy: 78.1 -> 79.3Improved
use RandAugment instead of SimAugmentPaCo ResNet-50 · ImageNet (full)top-1 accuracy: 78.7 -> 79.3Improved
add RandAugmentPaCo ResNet-152 · Places-LTtop-1 accuracy: 41.2 -> 41.2No clear effect

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