Research map / Label-efficient and robust learning / Self-supervised and contrastive
Mutual Contrastive Learning for Visual Representation Learning
Chuanguang Yang, Zhulin An, Linhang Cai, Yongjun Xu
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 |
|---|---|---|---|
| Add MCL with four-network collaborative training to supervised ResNet-32 | ResNet-32 · CIFAR-100 | Top-1 accuracy (%): 70.91$_{\pm 0.14}$ -> 74.04$_{\pm 0.07}$ | Improved |
| Increase MCL cohort size from two to four networks (ablation) | MCL with ResNet-32 · CIFAR-100 | Top-1 accuracy (%): 72.96$_{\pm 0.28}$ -> 74.04$_{\pm 0.07}$ | Improved |
| Add MCL with two-network training to MoCo | MoCo with ResNet-18 · ImageNet linear classification | Top-1 accuracy (%): 47.45$_{\pm 0.11}$ -> 48.04$_{\pm 0.13}$ | Improved |
| Add MCL with two-network training to MoCoV2 | MoCoV2 with ResNet-18 · ImageNet linear classification | Top-1 accuracy (%): 52.30$_{\pm 0.09}$ -> 52.76$_{\pm 0.06}$ | Improved |
| Add Soft VCL to VCL (ablation) | MCL with ResNet-32 · CIFAR-100 | Top-1 accuracy (%): 71.57$_{\pm 0.09}$ -> 73.06$_{\pm 0.15}$ | Improved |
| Add Soft ICL to ICL (ablation) | MCL with ResNet-32 · CIFAR-100 | Top-1 accuracy (%): 71.92$_{\pm 0.19}$ -> 73.68$_{\pm 0.13}$ | Improved |
| Use ICL and Soft ICL instead of VCL and Soft VCL (ablation) | MCL with ResNet-32 · CIFAR-100 | Top-1 accuracy (%): 73.06$_{\pm 0.15}$ -> 73.68$_{\pm 0.13}$ | Improved |
| Combine VCL, Soft VCL, ICL, and Soft ICL (ablation) | MCL with ResNet-32 · CIFAR-100 | Top-1 accuracy (%): 73.68$_{\pm 0.13}$ -> \textbf{74.04}$_{\pm 0.07}$ | Improved |
| Increase cohort size from five to six networks (ablation) | MCL with ResNet-32 · CIFAR-100 | Top-1 accuracy (%): \textbf{74.10}$_{\pm 0.10}$ -> 73.98$_{\pm 0.09}$ | Hurt |
| Increase cohort size from five to six networks (ablation) | MCL with WRN-16-2 · CIFAR-100 | Top-1 accuracy (%): \textbf{75.86}$_{\pm 0.25}$ -> 75.78$_{\pm 0.24}$ | Hurt |
Closest papers by meaning
- Online Knowledge Distillation via Mutual Contrastive Learning for Visual Recognition (TPAMI 2023 · similarity 0.93)
- A Broad Study on the Transferability of Visual Representations With Contrastive Learning (ICCV 2021 · similarity 0.90)
- A Simple Framework for Contrastive Learning of Visual Representations (ICML 2020 · similarity 0.89)
- Joint Contrastive Learning with Infinite Possibilities (NeurIPS 2020 · similarity 0.89)
- Prototypical Contrastive Learning of Unsupervised Representations (ICLR 2021 · similarity 0.88)
- Decoupled Contrastive Learning (ECCV 2022 · similarity 0.88)
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.
