Research map / Architectures and efficiency / Attention and transformers
MiniViT: Compressing Vision Transformers With Weight Multiplexing
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 weight transformation to weight sharing (ablation) | Mini-Swin-T · ImageNet-1K | Top-1 accuracy (%): 79.0 -> 79.2 | Helped |
| Add prediction-logit distillation to weight sharing (ablation) | Mini-Swin-T · ImageNet-1K | Top-1 accuracy (%): 79.0 -> 79.8 | Helped |
| Combine weight transformation and distillation with weight sharing (ablation) | Mini-Swin-T · ImageNet-1K | Top-1 accuracy (%): 79.0 -> 81.4 | Helped |
| Add weight transformation to weight sharing (ablation) | Mini-DeiT-S · ImageNet-1K | Top-1 accuracy (%): 78.3 -> 79.3 | Helped |
| Add prediction-logit distillation to weight sharing (ablation) | Mini-DeiT-S · ImageNet-1K | Top-1 accuracy (%): 78.3 -> 80.1 | Helped |
| Combine weight transformation and distillation with weight sharing (ablation) | Mini-DeiT-S · ImageNet-1K | Top-1 accuracy (%): 78.3 -> 80.7 | Helped |
| Add ground-truth labels to prediction-soft-label training (ablation) | Mini-Swin-T · ImageNet-1K | Top-1 accuracy (%): 81.2 -> 80.9 | Hurt |
| Add attention-level distillation to prediction-logit distillation (ablation) | Mini-Swin-T · ImageNet-1K | Top-1 accuracy (%): 81.2 -> 81.4 | Helped |
| Add hidden-state distillation to prediction-logit distillation (ablation) | Mini-Swin-T · ImageNet-1K | Top-1 accuracy (%): 81.2 -> 81.5 | Helped |
| Add attention-level and hidden-state distillation to prediction-logit distillation (ablation) | Mini-Swin-T · ImageNet-1K | Top-1 accuracy (%): 81.2 -> 81.4 | Helped |
| Share every two consecutive layers rather than all layers within stages (ablation) | Mini-Swin-T · ImageNet-1K | Top-1 accuracy (%): 81.4 -> 82.2 | Helped |
| Share every two consecutive layers rather than all layers (ablation) | Mini-DeiT-B · ImageNet-1K | Top-1 accuracy (%): 79.8 -> 83.2 | Helped |
Closest papers by meaning
- DeepCompress-ViT: Rethinking Model Compression to Enhance Efficiency of Vision Transformers at the Edge (CVPR 2025 · similarity 0.89)
- You Only Need Less Attention at Each Stage in Vision Transformers (CVPR 2024 · similarity 0.88)
- Bi-ViT: Pushing the Limit of Vision Transformer Quantization (AAAI 2024 · similarity 0.88)
- Dense Vision Transformer Compression with Few Samples (CVPR 2024 · similarity 0.88)
- RepViT: Revisiting Mobile CNN From ViT Perspective (CVPR 2024 · similarity 0.88)
- GradQ-ViT: Robust and Efficient Gradient Quantization for Vision Transformers (AAAI 2025 · similarity 0.88)
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.