Research map / Label-efficient and robust learning / Self-supervised and contrastive
Improvements to Context Based Self-Supervised Learning
Read the paperCheck 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 chroma blurring to the two-patch baseline (ablation) | Context-based self-supervised learning · CUB/CCars mean | classification accuracy: 71.09 -> 72.55 | Helped |
| Add yoked patch jitter (ablation) | Context-based self-supervised learning with chroma blurring · CUB/CCars mean | classification accuracy: 72.55 -> 73.06 | Helped |
| Add a third patch (ablation) | Context-based self-supervised learning with CB and YJ · CUB/CCars mean | classification accuracy: 73.06 -> 73.36 | Helped |
| Add extra patch configurations (ablation) | Context-based self-supervised learning with CB, YJ, and TP · CUB/CCars mean | classification accuracy: 73.36 -> 73.79 | Helped |
| Add usual augmentation tricks (ablation) | Context-based self-supervised learning with CB, YJ, TP, and EPC · CUB/CCars mean | classification accuracy: 73.79 -> 74.37 | Helped |
| Add random apertures to two of three patches (ablation) | Context-based self-supervised learning with CB, YJ, TP, EPC, and UBT · CUB/CCars mean | classification accuracy: 74.37 -> 75.04 | Helped |
| Add 180-degree rotation classification (ablation) | Context-based self-supervised learning with CB, YJ, TP, EPC, UBT, and RA · CUB/CCars mean | classification accuracy: 75.04 -> 76.56 | Helped |
| Add 90- and 270-degree rotation classification and weight varying (ablation) | Context-based self-supervised learning with 180-degree rotation classification · PASCAL VOC 2007 classification | mAP: 68.39 -> 68.31 | Hurt |
| Rotate patches without assigning separate rotation classes (ablation) | Context-based self-supervised learning with CB, YJ, TP, EPC, UBT, and RA · CUB/CCars mean | classification accuracy: 75.04 -> 75.04 | No clear effect |
| Add rotation classification rather than rotation-only augmentation (ablation) | Context-based self-supervised learning with CB, YJ, TP, EPC, UBT, and RA · CUB/CCars mean | classification accuracy: 75.04 -> 76.56 | Helped |
| Aperture all three patches instead of two (ablation) | Random Aperture · CUB/CCars mean | classification accuracy: 76.25 -> 75.49 | Hurt |
| Use hybrid patch configurations alone in the full-tool setting (ablation) | Context-based self-supervised learning with CB, YJ, TP, UBT, RA, and RWC · CUB/CCars mean | classification accuracy: 75.75 -> 74.83 | Hurt |
Closest papers by meaning
- How Useful Is Self-Supervised Pretraining for Visual Tasks? (CVPR 2020 · similarity 0.87)
- How Well Do Self-Supervised Models Transfer? (CVPR 2021 · similarity 0.87)
- Mix-and-Match Tuning for Self-Supervised Semantic Segmentation (AAAI 2018 · similarity 0.86)
- Colorization as a Proxy Task for Visual Understanding (CVPR 2017 · similarity 0.86)
- Intrinsic Self-Supervision for Data Quality Audits (NeurIPS 2024 · similarity 0.86)
- Probing the Mid-level Vision Capabilities of Self-Supervised Learning (CVPR 2025 · similarity 0.84)
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.