Research map / Label-efficient and robust learning / Recognition and long-tail
HCSC: Hierarchical Contrastive Selective Coding
Yuanfan Guo, Minghao Xu, Jiawen Li, Bingbing Ni, Xuanyu Zhu, Zhenbang Sun, Yi Xu
Check 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 hierarchical prototypes and selective instance-wise and prototypical contrastive learning | HCSC · ImageNet linear classification | Top1-Acc: 68.6 -> 69.2 | Improved |
| Add hierarchical prototypes to instance-wise and prototypical contrastive losses (ablation) | HCSC · ImageNet linear classification | Top1-Acc: 67.6 -> 68.1 | Improved |
| Add instance-wise pair selection (ablation) | HCSC · ImageNet linear classification | Top1-Acc: 68.1 -> 68.9 | Improved |
| Add prototypical pair selection alone (ablation) | HCSC · ImageNet linear classification | Top1-Acc: 68.1 -> 68.2 | Improved |
| Combine instance-wise and prototypical pair selection (ablation) | HCSC · ImageNet linear classification | Top1-Acc: 68.9 -> 69.2 | Improved |
| Add instance-wise contrastive loss (ablation) | HCSC · ImageNet linear classification | Top1-Acc: 65.7 -> 69.2 | Improved |
| Add prototypical contrastive loss (ablation) | HCSC · ImageNet linear classification | Top1-Acc: 68.2 -> 69.2 | Improved |
| Increase prototype hierarchy count from one to two (ablation) | HCSC · ImageNet linear classification | Top1-Acc: 68.0 -> 69.0 | Improved |
| Increase prototype hierarchy count from three to four (ablation) | HCSC · ImageNet linear classification | Top1-Acc: 69.2 -> 69.2 | No clear effect |
| Use insufficient prototype counts, 1000-500-200, rather than 3000-2000-1000 (ablation) | HCSC · ImageNet linear classification | Top1-Acc: 69.2 -> 68.7 | Hurt |
| Use initial learning rate 30.0 in downstream linear evaluation | HCSC · ImageNet linear classification | Top1-Acc: 67.5 -> 67.3 | Hurt |
| Pretrain with HCSC rather than AdCo | ImageNet-pretrained encoder · CUB zero-shot | KNN-Top1-Acc: 22.9 -> 26.9 | Improved |
Later papers that compared with this method
1 later accepted paper reports HCSC as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- HASSL: Hierarchy-Aware Self-Supervised Learning Framework for Single Cell Microscopy (ECCV 2026 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- AdCo* — AdCo: Adversarial Contrast for Efficient Learning of Unsupervised Representations From Self-Trained Negative Adversaries (CVPR 2021) · better on 6 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Seed the Views: Hierarchical Semantic Alignment for Contrastive Representation Learning (TPAMI 2022 · similarity 0.91)
- Variational Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.87)
- One Leaf Reveals the Season: Occlusion-Based Contrastive Learning with Semantic-Aware Views for Efficient Visual Representation (ICML 2025 · similarity 0.87)
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (ICLR 2025 · similarity 0.86)
- Diversity Is All You Need for Contrastive Learning: Spectral Bounds on Gradient Magnitudes (NeurIPS 2025 · similarity 0.86)
- On the Similarities of Embeddings in Contrastive Learning (ICML 2025 · similarity 0.86)
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.
