Research map / Label-efficient and robust learning / Recognition and long-tail
Iterative Reorganization With Weak Spatial Constraints: Solving Arbitrary Jigsaw Puzzles for Unsupervised 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 |
|---|---|---|---|
| Add binary spatial terms to unary terms (ablation) | Jigsaw pre-training with ResNet50 · PascalVOC 2007 | classification accuracy (%): 72.4 -> 75.3 | Helped |
| Add binary spatial terms to unary terms (ablation) | Jigsaw pre-training with ResNet18 · PascalVOC 2007 | classification accuracy (%): 72.5 -> 74.7 | Helped |
| Add mirror augmentation (ablation) | Jigsaw pre-training with AlexNet · PascalVOC 2007 | classification accuracy (%): 66.6 -> 68.1 | Helped |
| Add mirror augmentation (ablation) | Jigsaw pre-training with AlexNet · ILSVRC2012 validation | puzzle recognition accuracy (%): 32.2 -> 3.4 | Hurt |
| Add mirror augmentation (ablation) | Jigsaw pre-training with ResNet50 · PascalVOC 2007 | classification accuracy (%): 72.4 -> 73.1 | Helped |
| Add mirror augmentation (ablation) | Jigsaw pre-training with ResNet50 · ILSVRC2012 validation | puzzle recognition accuracy (%): 47.3 -> 4.9 | Hurt |
| Use 3×3×3 rather than 2×2×2 puzzles (ablation) | VNet initialized by jigsaw pre-training on MSD · NIH pancreas segmentation, 10% training data | DSC (%): $68.44$ -> $72.24$ | Helped |
| Use 3×3×3 rather than 2×2×2 puzzles (ablation) | VNet initialized by jigsaw pre-training on MSD · NIH pancreas segmentation, 20% training data | DSC (%): $76.58$ -> $77.80$ | Helped |
| Use 3×3×3 rather than 2×2×2 puzzles (ablation) | VNet initialized by jigsaw pre-training on MSD · NIH pancreas segmentation, 100% training data | DSC (%): $81.48$ -> $82.33$ | Helped |
| Pre-train on MSD rather than NIH (ablation) | VNet with 3×3×3 jigsaw pre-training · NIH pancreas segmentation, 10% training data | DSC (%): $70.80$ -> $72.24$ | Helped |
| Pre-train on MSD rather than NIH (ablation) | VNet with 3×3×3 jigsaw pre-training · NIH pancreas segmentation, 20% training data | DSC (%): $76.50$ -> $77.80$ | Helped |
| Pre-train on MSD rather than NIH (ablation) | VNet with 3×3×3 jigsaw pre-training · NIH pancreas segmentation, 100% training data | DSC (%): $81.68$ -> $82.33$ | Helped |
Closest papers by meaning
- Domain Generalization by Solving Jigsaw Puzzles (CVPR 2019 · similarity 0.88)
- Solving Mixed-Modal Jigsaw Puzzle for Fine-Grained Sketch-Based Image Retrieval (CVPR 2020 · similarity 0.88)
- Solving Masked Jigsaw Puzzles with Diffusion Vision Transformers (CVPR 2024 · similarity 0.87)
- Sliding Puzzles Gym: A Scalable Benchmark for State Representation in Visual Reinforcement Learning (ICML 2025 · similarity 0.83)
- SPTNet: An Efficient Alternative Framework for Generalized Category Discovery with Spatial Prompt Tuning (ICLR 2024 · similarity 0.83)
- Visual Jigsaw Post-Training Improves MLLMs (ICLR 2026 · similarity 0.83)
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.