Research map / Detection and segmentation / Open-set and few-shot detection
Unsupervised Learning of Visual Representations by Solving Jigsaw Puzzles
Mehdi Noroozi, Paolo Favaro
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 a random gap between tiles (ablation) | CFN Jigsaw · PASCAL VOC 2007 detection (Jigsaw ablation) | mAP: 47.7 -> 52.6 | Improved |
| add per-patch mean/std normalization (ablation) | CFN Jigsaw · PASCAL VOC 2007 detection (Jigsaw ablation) | mAP: 43.5 -> 52.6 | Improved |
| add color channel jittering (ablation) | CFN Jigsaw · PASCAL VOC 2007 detection (Jigsaw ablation) | mAP: 51.1 -> 52.6 | Improved |
| remove the gap shortcut guard: jigsaw task becomes easier (ablation) | CFN Jigsaw · Jigsaw task accuracy | accuracy: 88 -> 98 | Mixed |
| raise average Hamming distance of permutation set from 3.99 to 8.00 (ablation) | CFN Jigsaw · PASCAL VOC 2007 detection (Jigsaw ablation) | mAP: 50.2 -> 53.2 | Improved |
| raise average Hamming distance of permutation set from 6.35 to 8.00 (ablation) | CFN Jigsaw · PASCAL VOC 2007 detection (Jigsaw ablation) | mAP: 51.3 -> 53.2 | Improved |
| use 1000 permutations instead of 10 (ablation) | CFN Jigsaw · PASCAL VOC 2007 detection (Jigsaw ablation) | mAP: 49.2 -> 53.2 | Improved |
| use 1000 permutations instead of 6 (easier, unambiguous task) (ablation) | CFN Jigsaw · PASCAL VOC 2007 detection (Jigsaw ablation) | mAP: 49.7 -> 53.2 | Improved |
| raise minimum Hamming distance from 2 to 5 with 71 permutations (ablation) | CFN Jigsaw · PASCAL VOC 2007 detection (Jigsaw ablation) | mAP: 52.6 -> 52.8 | No clear effect |
| use Jigsaw CFN pretraining instead of context prediction by Pathak et al. | AlexNet · PASCAL VOC 2007 detection | mAP: 44.5 -> 53.2 | Improved |
| use Jigsaw CFN pretraining instead of Doersch et al. | AlexNet · PASCAL VOC 2007 detection | mAP: 46.6 -> 53.2 | Improved |
| use Jigsaw CFN pretraining instead of Pathak et al. | AlexNet · PASCAL VOC 2012 segmentation | mIoU: 29.7 -> 37.6 | Improved |
Closest papers by meaning
- Iterative Reorganization With Weak Spatial Constraints: Solving Arbitrary Jigsaw Puzzles for Unsupervised Representation Learning (CVPR 2019 · similarity 0.91)
- Jigsaw Clustering for Unsupervised Visual Representation Learning (CVPR 2021 · similarity 0.88)
- Solving Masked Jigsaw Puzzles with Diffusion Vision Transformers (CVPR 2024 · similarity 0.86)
- Towards flexible perception with visual memory (ICML 2025 · similarity 0.86)
- Factorized Diffusion Autoencoder for Unsupervised Disentangled Representation Learning (AAAI 2024 · similarity 0.85)
- PICNN: A Pathway towards Interpretable Convolutional Neural Networks (AAAI 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.
