Research map / Label-efficient and robust learning / Recognition and long-tail
Relative Contrastive Loss for Unsupervised Representation Learning
Shixiang Tang, Feng Zhu, Lei Bai, Rui Zhao, Wanli Ouyang
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 |
|---|---|---|---|
| increase hierarchy levels from 1 to 2 (ablation) | RCL · ImageNet-1K linear eval (200 epochs, predictor/level ablation) | top-1 accuracy: 70.2 -> 71.3 | Improved |
| increase hierarchy levels from 2 to 3 (ablation) | RCL · ImageNet-1K linear eval (200 epochs, predictor/level ablation) | top-1 accuracy: 71.3 -> 71.8 | Improved |
| increase hierarchy levels from 3 to 4 (ablation) | RCL · ImageNet-1K linear eval (200 epochs, predictor/level ablation) | top-1 accuracy: 71.8 -> 71.4 | Hurt |
| use multiple criterion-specific predictors instead of one (3 levels) (ablation) | RCL · ImageNet-1K linear eval (200 epochs, predictor/level ablation) | top-1 accuracy: 71.8 -> 72.6 | Improved |
| replace NNCLR nearest-neighbor positives with offline K-means relative labels (ablation) | RCL · ImageNet-1K linear eval (200 epochs, clustering ablation) | top-1 accuracy: 70.7 -> 71.8 | Improved |
| use hierarchical DBSCAN labels instead of K-means (ablation) | RCL · ImageNet-1K linear eval (200 epochs, clustering ablation) | top-1 accuracy: 71.8 -> 72.3 | Improved |
| use online hierarchical clustering instead of offline DBSCAN (ablation) | RCL · ImageNet-1K linear eval (200 epochs, clustering ablation) | top-1 accuracy: 72.3 -> 72.6 | Improved |
| online hierarchical clustering vs offline DBSCAN epoch time (ablation) | RCL · ImageNet-1K linear eval (200 epochs, clustering ablation) | training time per epoch: 986s -> 776s | Improved |
| increase support queue size from 65536 to 98304 (ablation) | RCL · ImageNet-1K linear eval (200 epochs) | top-1 accuracy: +0.36% | Improved |
| increase support queue beyond 98304 (ablation) | RCL · ImageNet-1K linear eval (200 epochs) | top-1 accuracy | No clear effect |
| RCL vs MoCo v3 (800 epochs) | RCL · ImageNet-1K linear eval (800+ epochs) | top-1 accuracy: 73.8 -> 75.8 | Improved |
| RCL vs NNCLR (200 epochs) | RCL · ImageNet-1K linear eval (200 epochs) | top-1 accuracy: 70.7 -> 72.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Barlow Twins — Barlow Twins: Self-Supervised Learning via Redundancy Reduction (ICML 2021) · better on 2 of 2 shared results
- Adco — AdCo: Adversarial Contrast for Efficient Learning of Unsupervised Representations From Self-Trained Negative Adversaries (CVPR 2021) · better on 1 of 1 shared result
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
- Conditional Negative Sampling for Contrastive Learning of Visual Representations (ICLR 2021 · similarity 0.90)
- Robust Contrastive Learning Using Negative Samples with Diminished Semantics (NeurIPS 2021 · similarity 0.89)
- CaCo: Both Positive and Negative Samples are Directly Learnable via Cooperative-Adversarial Contrastive Learning (TPAMI 2023 · similarity 0.88)
- On the Similarities of Embeddings in Contrastive Learning (ICML 2025 · similarity 0.88)
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (ICLR 2025 · similarity 0.87)
- Decoupled Contrastive Learning for Long-Tailed Recognition (AAAI 2024 · 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.
