Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-Supervised Representation Learning by Rotation Feature Decoupling
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 |
|---|---|---|---|
| Add instance discrimination to rotation prediction (ablation) | Rotation prediction representation · ImageNet ILSVRC 2012 val, conv5 | top-1 accuracy: 36.5 -> 44.0 | Helped |
| Add PU weighting to reduce the influence of noisy rotation labels (ablation) | Rotation + Instance · ImageNet ILSVRC 2012 val, conv5 | top-1 accuracy: 44.0 -> 44.3 | Helped |
| Use rotation prediction only, without instance discrimination (ablation) | Self-supervised feature learning · ImageNet ILSVRC 2012 val, conv5 | top-1 accuracy: 44.3 -> 36.5 | Hurt |
| Use instance discrimination only, without rotation prediction (ablation) | Self-supervised feature learning · ImageNet ILSVRC 2012 val, conv5 | top-1 accuracy: 44.3 -> 32.9 | Hurt |
| Move feature decoupling from conv5 to fc6 (ablation) | Full model · ImageNet ILSVRC 2012 val, conv5 | top-1 accuracy: 41.0 -> 42.6 | Helped |
| Move feature decoupling from fc6 to fc7 (ablation) | Full model · ImageNet ILSVRC 2012 val, conv5 | top-1 accuracy: 42.6 -> 44.3 | Helped |
| Use only the rotation-unrelated feature half instead of the full decoupled feature (ablation) | Ours · Rotated PASCAL VOC classification, fc8 | classification accuracy: 59.2 -> 57.7 | Hurt |
| Add the rotation-related feature half to the rotation-unrelated half (ablation) | Ours · Rotated PASCAL VOC classification, fc8 | classification accuracy: 57.7 -> 59.2 | Helped |
| Use only rotation-related features rather than the decoupled feature (ablation) | Ours · Rotated PASCAL VOC classification, fc8 | classification accuracy: 59.2 -> 38.6 | Hurt |
| Use rotation-decoupled features rather than RotNet features | ImageNet-pretrained self-supervised representation · ImageNet ILSVRC 2012 val, conv5 | top-1 accuracy: 36.5 -> 44.3 | Helped |
| Use rotation-decoupled features rather than RotNet features | ImageNet-pretrained self-supervised representation · Places val, conv5 | top-1 accuracy: 33.7 -> 38.6 | Helped |
| Use rotation-decoupled features rather than RotNet features | PASCAL VOC 2007 transfer · Classification, fc6-8 | mAP: 73.0 -> 74.3 | Helped |
Closest papers by meaning
- Revisiting Self-Supervised Visual Representation Learning (CVPR 2019 · similarity 0.89)
- Unsupervised Representation Learning by Predicting Image Rotations (ICLR 2018 · similarity 0.89)
- Learning with 3D rotations, a hitchhiker's guide to SO(3) (ICML 2024 · similarity 0.88)
- Exploring the Equivalence of Siamese Self-Supervised Learning via a Unified Gradient Framework (CVPR 2022 · similarity 0.87)
- Learning predictable and robust neural representations by straightening image sequences (NeurIPS 2024 · similarity 0.85)
- Structuring Representation Geometry with Rotationally Equivariant Contrastive Learning (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.