Research map / Architectures and efficiency / Attention and transformers
Masked Jigsaw Puzzle: A Versatile Position Embedding for Vision Transformers
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 MJP to DeiT-S | DeiT-S · ImageNet-1K | Top-1 Acc.: 79.8 -> 80.5 | Helped |
| Add MJP to DeiT-S | DeiT-S · ImageNet-1K | Consistency: 64.3 -> 82.9 | Helped |
| Add MJP to Swin-T | Swin-T · ImageNet-1K | Consistency: 41.5 -> 66.9 | Helped |
| Add MJP to Swin-T | Swin-T · ImageNet-1K | Top-1 Acc.: 81.3 -> 81.3 | No clear effect |
| Add MJP to DeiT-S | DeiT-S · ImageNet-C | mCE: 54.6 -> 40.78 | Helped |
| Add MJP to DeiT-S | DeiT-S · ImageNet-A | Acc: 19.2 -> 21.6 | Helped |
| Add MJP to DeiT-S | DeiT-S · ImageNet-O | AUPR: 20.9 -> 22.6 | Helped |
| Train with MJP | DeiT-S · ImageNet-1K validation, analytic gradient attack setting a | MSE: .0350 -> .1055 | Helped |
| Train with MJP | DeiT-S · ImageNet-1K validation, analytic gradient attack setting a | LPIPS: .3767 -> .6545 | Helped |
| Increase the MJP masking ratio from 0 to 0.27 (ablation) | DeiT-S + MJP · ImageNet-1K | Consistency: 64.0 -> 93.1 | Helped |
| Increase the MJP masking ratio from 0 to 0.27 (ablation) | DeiT-S + MJP · ImageNet-1K | Top-1 Acc.: 80.0 -> 80.3 | Mixed |
| Shuffle patches without unknown embeddings or DAL (ablation) | DeiT-S · ImageNet-1K | Top-1 Acc.: 79.8 -> 79.2 | Hurt |
Closest papers by meaning
- Solving Masked Jigsaw Puzzles with Diffusion Vision Transformers (CVPR 2024 · similarity 0.87)
- Maximizing the Position Embedding for Vision Transformers with Global Average Pooling (AAAI 2025 · similarity 0.87)
- Gaussian Splatting Feature Fields for (Privacy-Preserving) Visual Localization (CVPR 2025 · similarity 0.85)
- Stochastic positional embeddings improve masked image modeling (ICML 2024 · similarity 0.84)
- Configuring Data Augmentations to Reduce Variance Shift in Positional Embedding of Vision Transformers (AAAI 2025 · similarity 0.84)
- Circle-RoPE: Cone-like Decoupled Rotary Positional Embedding for Vision-Language Models (ICML 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.