Research map / Label-efficient and robust learning / Recognition and long-tail
Instance Similarity Learning for Unsupervised Feature Representation
Ziwei Wang, Yunsong Wang, Ziyi Wu, Jiwen Lu, Jie Zhou
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 hard positive enhancement (ablation) | ISL AlexNet · CIFAR-10 (weighted kNN, FC) | top-1 accuracy (%): 81.1 -> 82.8 | Improved |
| add hard positive enhancement (ablation) | ISL ResNet18 · CIFAR-100 (weighted kNN, FC) | top-1 accuracy (%): 52.1 -> 54.7 | Improved |
| add hard positive enhancement (ablation) | ISL ResNet50 · CIFAR-100 (weighted kNN, FC) | top-1 accuracy (%): 56.7 -> 58.1 | Improved |
| add hard positive enhancement (ablation) | ISL ResNet50 · CIFAR-100 (linear, conv5) | top-1 accuracy (%): 63.0 -> 65.9 | Improved |
| add hard positive enhancement (ablation) | ISL AlexNet · SVHN (weighted kNN, FC) | top-1 accuracy (%): 91.0 -> 91.8 | Improved |
| use instance similarity learning without HPE instead of AND neighborhood discovery | ISL w/o HPE AlexNet · CIFAR-100 (weighted kNN, FC) | top-1 accuracy (%): 41.5 -> 49.2 | Improved |
| use ISL instead of AND | ResNet50 · CIFAR-100 (weighted kNN, FC) | top-1 accuracy (%): 49.0 -> 58.1 | Improved |
| use ISL instead of RotNet | AlexNet · CIFAR-10 (linear, conv5) | top-1 accuracy (%): 84.1 -> 85.8 | Improved |
| use ISL instead of LA | ResNet50 · ImageNet (weighted kNN, FC) | top-1 accuracy (%): 49.4 -> 50.2 | Improved |
| integrate ISL with MoCo-v2 | MoCo-v2 ResNet50 · ImageNet (linear, conv5) | top-1 accuracy (%): 67.5 -> 68.6 | Improved |
| integrate ISL with MoCo-v2 | MoCo-v2 ResNet50 · ImageNet (weighted kNN, FC) | top-1 accuracy (%): 38.5 -> 40.1 | Improved |
| integrate ISL with MoCo-v2 | MoCo-v2 ResNet50 · ImageNet (linear, conv4) | top-1 accuracy (%): 52.9 -> 51.7 | Hurt |
Closest papers by meaning
- Unsupervised Feature Learning via Non-Parametric Instance Discrimination (CVPR 2018 · similarity 0.89)
- Deep Unsupervised Similarity Learning Using Partially Ordered Sets (CVPR 2017 · similarity 0.88)
- Train a One-Million-Way Instance Classifier for Unsupervised Visual Representation Learning (AAAI 2021 · similarity 0.87)
- Unsupervised Domain Adaptation With Similarity Learning (CVPR 2018 · similarity 0.86)
- Unsupervised Feature Learning by Cross-Level Instance-Group Discrimination (CVPR 2021 · similarity 0.86)
- Mining on Manifolds: Metric Learning Without Labels (CVPR 2018 · 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.
