Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-Supervised Learning Across Domains
Silvia Bucci, Antonio D’Innocente, Yujun Liao, Fabio Maria Carlucci, Barbara Caputo B, Tatiana Tommasi
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 jigsaw self-supervised task to supervised training (ablation) | AlexNet DeepAll · PACS DG | accuracy: 71.81 -> 73.05 | Improved |
| add rotation self-supervised task to supervised training (ablation) | ResNet-18 DeepAll · PACS DG | accuracy: 78.40 -> 81.87 | Improved |
| combine jigsaw and rotation instead of jigsaw alone (ablation) | AlexNet multi-task · PACS DG | accuracy: 73.05 -> 74.08 | Improved |
| combine jigsaw and rotation instead of rotation alone (ablation) | ResNet-18 multi-task · PACS DG | accuracy: 81.87 -> 81.41 | No clear effect |
| add rotation task to supervised training (ablation) | AlexNet DeepAll · VLCS DG | accuracy: 72.49 -> 73.88 | Improved |
| combine jigsaw and rotation instead of rotation alone (ablation) | AlexNet multi-task · VLCS DG | accuracy: 73.88 -> 73.15 | Hurt |
| use rotation instead of jigsaw as auxiliary task (ablation) | ResNet-18 multi-task · Office-Home DG | accuracy: 61.20 -> 63.53 | Improved |
| use whole-image jigsaw pretext (J-AlexNet) instead of patch-based J-CFN (ablation) | self-supervised pretraining · PACS DG | accuracy: 59.41 -> 51.48 | Hurt |
| add jigsaw auxiliary task to the multi-branch CFN architecture (ablation) | C-CFN DeepAll · PACS DG | accuracy: 62.66 -> 64.89 | Improved |
| add self-supervised task on target data (ablation) | Jigsaw+Rotation ResNet-18 · PACS multi-source DA | accuracy: 83.73 -> 88.66 | Improved |
| add entropy loss (ablation) | Jigsaw+Rotation ResNet-18 · PACS multi-source DA | accuracy: 85.23 -> 88.66 | Improved |
| add self-supervised task on source data (ablation) | Jigsaw ResNet-50 · Office-Home single-source DA | accuracy: 62.06 -> 62.17 | No clear effect |
Closest papers by meaning
- Domain Generalization by Solving Jigsaw Puzzles (CVPR 2019 · similarity 0.92)
- Boosting Self-Supervised Learning via Knowledge Transfer (CVPR 2018 · similarity 0.90)
- Learning Across Tasks and Domains (ICCV 2019 · similarity 0.89)
- Self-ensembling for visual domain adaptation (ICLR 2018 · similarity 0.89)
- A General Representation-Based Approach to Multi-Source Domain Adaptation (ICML 2025 · similarity 0.88)
- Large Self-Supervised Models Bridge the Gap in Domain Adaptive Object Detection (CVPR 2025 · similarity 0.87)
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.
