AI Paper Atlas

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

Unleashing the Power of Contrastive Self-Supervised Visual Models via Contrast-Regularized Fine-Tuning

Yifan Zhang, Bryan Hooi, Dapeng Hu, Jian Liang, Jiashi Feng

NeurIPS 2021 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Contrastive versus non-contrastive theory · 31 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 Core-tuning instead of CE-tuningMoCo-v2 ResNet-50 fine-tuning · Image classification, 9-dataset Avg.top-1 accuracy: 87.76 -> 90.47Improved
Use Core-tuning instead of CE-tuningMoCo-v2 ResNet-50 fine-tuning · iNaturalist18top-1 accuracy: 61.72+/-0.18 -> 63.57+/-0.09Improved
Use Core-tuning instead of CE fine-tuningMoCo-v2 ResNet-50 semantic segmentation · PASCAL VOC val2012MIoU: 78.42 -> 79.62Improved
Use Core-tuning instead of CE-ConMoCo-v2 ResNet-50 domain generalization · PACS Avg.accuracy: 85.65 -> 88.08Improved
Use Core-tuning instead of CE-ConMoCo-v2 ResNet-50 domain generalization · VLCS Avg.accuracy: 77.67 -> 80.34Improved
Use Core-tuning adversarial training instead of AT-CE-ConMoCo-v2 ResNet-50 · CIFAR10, PGD-10 ℓ2 attack ε=0.5robust accuracy: 90.74 -> 92.97Improved
Add the original contrastive loss to CE-only fine-tuning (ablation)MoCo-v2 ResNet-50 · ImageNet20top-1 accuracy: 88.28 -> 89.29Improved
Use manifold mixup instead of CE-only fine-tuning (ablation)MoCo-v2 ResNet-50 · ImageNet20top-1 accuracy: 88.28 -> 90.67Improved
Use hardness-directed mixup with original contrastive loss instead of manifold mixup alone (ablation)MoCo-v2 ResNet-50 · ImageNet20top-1 accuracy: 90.67 -> 92.20Improved
Replace original contrastive loss with focal contrastive loss while retaining hardness-directed mixup (ablation)MoCo-v2 ResNet-50 · ImageNet20top-1 accuracy: 92.20 -> 92.73Improved
Add Core-tuning to fine-tuning across the nine image-classification datasetsResNet-50 with contrastive self-supervised pre-training · Image classification, 9 natural image datasetstop-1 accuracyImproved

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