Research map / Label-efficient and robust learning / Recognition and long-tail
Generalized Parametric Contrastive Learning
Jiequan Cui, Zhisheng Zhong, Zhuotao Tian, Shu Liu, Bei Yu, Jiaya Jia
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 changed | Where | Measured | Effect |
|---|---|---|---|
| use GPaCo instead of PaCo (no momentum encoder, learnable-center generalization) | PaCo · ImageNet-LT (ResNet-50) | top-1 accuracy: 57.0 -> 58.5 | Improved |
| use GPaCo instead of PaCo | PaCo · iNaturalist 2018 (ResNet-50) | top-1 accuracy: 73.2 -> 75.4 | Improved |
| use GPaCo instead of PaCo | PaCo · full ImageNet (ResNet-50) | top-1 accuracy: 79.3 -> 79.7 | Improved |
| use GPaCo instead of Supcon | Supcon · full CIFAR-100 | top-1 accuracy: 76.5 -> 80.3 | Improved |
| add GPaCo as auxiliary loss for semantic segmentation | UperNet Swin-T · ADE20K | mIoU (s.s.): 44.5 -> 45.4 | Improved |
| use GPaCo fine-tuning instead of MAE | MAE ViT-B · full ImageNet | top-1 accuracy: 82.5 -> 84.0 | Improved |
| use two views instead of a single view (ablation) | GPaCo ViT-B · full ImageNet | top-1 accuracy: 82.5 -> 84.0 | Improved |
| use random resized crop scale (0.2,1) instead of (0.08,1) (ablation) | PaCo · ImageNet-LT (ResNet-50) | top-1 accuracy: 55.0 -> 52.2 | Hurt |
| use RandAug for view 1 instead of SimAug (ablation) | PaCo · ImageNet-LT (ResNet-50) | top-1 accuracy: 55.0 -> 57.0 | Improved |
| use RandAug for both views instead of one (ablation) | PaCo · ImageNet-LT (ResNet-50) | top-1 accuracy: 57.0 -> 56.5 | Hurt |
| use stronger augmentation (RandAugStack) for view 2 instead of SimAug (ablation) | GPaCo · ImageNet-LT (ResNet-50) | top-1 accuracy: 57.9 -> 58.5 | Improved |
| use PaCo instead of multi-task CE+re-weighting (ablation) | CE · ImageNet-LT (ResNet-50) | top-1 accuracy: 49.0 -> 51.0 | Improved |
Closest papers by meaning
- Parametric Contrastive Learning (ICCV 2021 · similarity 0.93)
- Probabilistic Contrastive Learning for Long-Tailed Visual Recognition (TPAMI 2024 · similarity 0.86)
- Continuous Contrastive Learning for Long-Tailed Semi-Supervised Recognition (NeurIPS 2024 · similarity 0.84)
- Long-tailed Object Detection Pretraining: Dynamic Rebalancing Contrastive Learning with Dual Reconstruction (NeurIPS 2024 · similarity 0.84)
- Generalization Analysis for Supervised Contrastive Representation Learning under Non-IID Settings (ICML 2025 · similarity 0.82)
- A Tale of Two Classes: Adapting Supervised Contrastive Learning to Binary Imbalanced Datasets (CVPR 2025 · similarity 0.82)
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.
