Research map / Architectures and efficiency / Attention and transformers
SimMIM: A Simple Framework for Masked Image Modeling
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 |
|---|---|---|---|
| Use SimMIM pre-training instead of the previous best listed approach | ViT-B · ImageNet-1K | fine-tuning top-1 accuracy: +0.6% | Helped |
| Use SimMIM pre-training instead of supervised training | SwinV2-H · ImageNet-1K | top-1 accuracy: 83.3 -> 87.1 | Helped |
| Use random masking with patch size 32 rather than block-wise masking (ablation) | Swin-B · ImageNet-1K | fine-tuning top-1 accuracy: 82.7 -> 83.0 | Helped |
| Increase mask ratio from 0.4 to 0.8 with random patch size 4, 8, or 16 (ablation) | Swin-B SimMIM · ImageNet-1K | fine-tuning top-1 accuracy: +0.2%; +0.4%; +0.4% | Helped |
| Increase random masked patch size from 32 to 64 (ablation) | Swin-B SimMIM · ImageNet-1K | fine-tuning top-1 accuracy: 82.8 -> 82.6 | Hurt |
| Use a linear prediction head instead of an inverse Swin-T or inverse Swin-B (ablation) | Swin-B SimMIM · ImageNet-1K | fine-tuning top-1 accuracy: 82.4 -> 82.8; 82.5 -> 82.8 | Helped |
| Use a linear prediction head rather than a 2-layer MLP (ablation) | Swin-B SimMIM · ImageNet-1K | fine-tuning top-1 accuracy: 82.8 -> 82.8 | No clear effect |
| Reduce target resolution to 6^2 (ablation) | Swin-B SimMIM · ImageNet-1K | fine-tuning top-1 accuracy: 82.8 -> 82.3 | Hurt |
| Predict the full image, including visible pixels, rather than masked pixels only (ablation) | Swin-B SimMIM · ImageNet-1K | fine-tuning top-1 accuracy: 82.8 -> 81.7 | Hurt |
| Replace raw-pixel L1 regression with BEiT token classification (ablation) | Swin-B SimMIM · ImageNet-1K | fine-tuning top-1 accuracy: 82.8 -> 82.7 | No clear effect |
| Replace raw-pixel L1 regression with 8-bin color discretization (ablation) | Swin-B SimMIM · ImageNet-1K | fine-tuning top-1 accuracy: 82.8 -> 82.7 | No clear effect |
| Replace L1 with L2 or smooth-L1 regression (ablation) | Swin-B SimMIM · ImageNet-1K | fine-tuning top-1 accuracy: 82.8 -> 82.7 | No clear effect |
Closest papers by meaning
- From Prototypes to General Distributions: An Efficient Curriculum for Masked Image Modeling (CVPR 2025 · similarity 0.89)
- Morphing Tokens Draw Strong Masked Image Models (ICLR 2025 · similarity 0.88)
- On the Role of Discrete Tokenization in Visual Representation Learning (ICLR 2024 · similarity 0.87)
- Wavelet-Driven Masked Image Modeling: A Path to Efficient Visual Representation (AAAI 2025 · similarity 0.87)
- Pre-training with Random Orthogonal Projection Image Modeling (ICLR 2024 · similarity 0.87)
- Patch-Aware Sample Selection for Efficient Masked Image Modeling (AAAI 2024 · 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.