Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-Supervised Learning of Pretext-Invariant Representations
Ishan Misra, Laurens van der Maaten
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 |
|---|---|---|---|
| Learn representations invariant to Jigsaw transformations rather than predicting patch permutations | Jigsaw self-supervised learning · VOC07 test (VOC07+12 train) | APall: 48.9 -> 54.0 | Improved |
| Learn representations invariant to Jigsaw transformations rather than predicting patch permutations | Jigsaw self-supervised learning · ImageNet | top-1 accuracy: 45.7 -> 63.6 | Improved |
| Use PIRL instead of supervised pretraining | ResNet-50 Faster R-CNN · VOC07 test (VOC07+12 train) | APall: 52.6 -> 54.0 | Improved |
| Set the PIRL loss mixture to λ=0.5 instead of λ=0 (NPID++) (ablation) | PIRL · ImageNet | top-1 accuracy | Improved |
| Vary λ away from 0.5 (ablation) | PIRL · ImageNet | top-1 accuracy | Mixed |
| Increase the number of Jigsaw patch permutations (ablation) | PIRL · VOC07 | mAP | Improved |
| Increase the number of negative samples (ablation) | PIRL · ImageNet | top-1 accuracy | Improved |
| Use rotation-invariant PIRL rather than the covariant Rotation task | Rotation pretext learning · ImageNet | top-1 accuracy: +11.3 | Improved |
| Combine rotation with Jigsaw transforms (ablation) | PIRL (Jigsaw) · ImageNet | top-1 accuracy: 62.2 -> 63.1 | Improved |
| Pretrain on YFCC-1M rather than ImageNet | PIRL · ImageNet | top-1 accuracy: 63.6 -> 57.8 | Hurt |
| Pretrain on YFCC-1M rather than ImageNet | PIRL · Places205 | top-1 accuracy: 49.8 -> 51.0 | Improved |
| Double the number of ResNet-50 channels (ablation) | PIRL · ImageNet | top-1 accuracy: 63.6 -> 67.4 | Improved |
Closest papers by meaning
- Learning Representations by Predicting Bags of Visual Words (CVPR 2020 · similarity 0.87)
- Equivariant Self-Supervised Learning: Encouraging Equivariance in Representations (ICLR 2022 · similarity 0.87)
- Self-supervised Learning from a Multi-view Perspective (ICLR 2021 · similarity 0.87)
- Predicting What You Already Know Helps: Provable Self-Supervised Learning (NeurIPS 2021 · similarity 0.87)
- In-Context Symmetries: Self-Supervised Learning through Contextual World Models (NeurIPS 2024 · similarity 0.86)
- Self-Organizing Visual Prototypes for Non-Parametric Representation Learning (ICML 2025 · 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.
