Research map / Label-efficient and robust learning / Self-supervised and contrastive
Jigsaw Clustering for Unsupervised Visual Representation Learning
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 Jigsaw Clustering rather than the strongest single-batch baseline | ImageNet-1k linear evaluation · ImageNet-1k | accuracy: 63.8 -> 66.4 | Helped |
| Use Jigsaw Clustering rather than MoCo v2 | ImageNet-1k linear evaluation · ImageNet-1k | accuracy: 66.4 vs 67.7 | Hurt |
| Use Jigsaw Clustering rather than MoCo v2 | linear evaluation on smaller training sets · ImageNet-100 and ImageNet-10% | accuracy: 74.7 -> 80.9; 38.3 -> 44.3 | Helped |
| Use JigClu pretrained weights rather than MoCo v2 weights | semi-supervised ImageNet classification · ImageNet-1k 1% and 10% labels | Top-1 accuracy: 34.5 -> 40.7; 61.1 -> 63.0 | Helped |
| Use JigClu pretrained weights rather than MoCo v2 weights | COCO detection · COCO detection | AP: 38.9 -> 39.3 | Helped |
| Use JigClu pretrained weights rather than supervised ImageNet weights | CIFAR finetuning · CIFAR-10 and CIFAR-100 | accuracy: 88.6 -> 89.5; 60.6 -> 64.7 | Helped |
| Use montage images instead of small-size patches (ablation) | ImageNet-100 linear evaluation · ImageNet-100 | accuracy: 67.0 -> 70.9 | Helped |
| Use montage images instead of scaled-up patches (ablation) | ImageNet-100 linear evaluation · ImageNet-100 | accuracy: 71.3 -> 70.9 | Hurt |
| Apply augmentation to each patch after splitting, rather than before splitting (ablation) | JigClu · ImageNet-100 | linear-evaluation accuracy: 3.7 -> 80.9 | Helped |
| Increase the patch grid from m=2 to m=3 or m=4 (ablation) | JigClu · ImageNet-100 | linear-evaluation accuracy: 80.9 -> 74.7; 80.9 -> 70.1 | Hurt |
| Add the location branch to the clustering branch (ablation) | JigClu · ImageNet-1k | linear-evaluation accuracy: 65.1 -> 66.4 | Helped |
| Add the clustering branch to the location branch (ablation) | JigClu · ImageNet-1k | linear-evaluation accuracy: 3.2 -> 66.4 | Helped |
Closest papers by meaning
- Unsupervised Visual Representation Learning by Online Constrained K-Means (CVPR 2022 · similarity 0.89)
- Beyond Supervised vs. Unsupervised: Representative Benchmarking and Analysis of Image Representation Learning (CVPR 2022 · similarity 0.89)
- Iterative Reorganization With Weak Spatial Constraints: Solving Arbitrary Jigsaw Puzzles for Unsupervised Representation Learning (CVPR 2019 · similarity 0.87)
- ClusterFit: Improving Generalization of Visual Representations (CVPR 2020 · similarity 0.87)
- Unsupervised Learning of Visual Features by Contrasting Cluster Assignments (NeurIPS 2020 · similarity 0.86)
- Image Clustering via the Principle of Rate Reduction in the Age of Pretrained Models (ICLR 2024 · similarity 0.85)
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.