Research map / Architectures and efficiency / Attention and transformers
Width & Depth Pruning 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 |
|---|---|---|---|
| Apply WDPruning width ratio 0.2 while retaining 12 blocks | DeiT-Base · ILSVRC-12 | Top-1 accuracy (%): 81.80 -> 81.94 (+0.14) | Helped |
| Add learned layer-wise pruning ratios instead of uniform pruning (ablation) | DeiT-Small · CIFAR-10 | Top-1 accuracy (%): 75.26 -> 78.55 | Helped |
| Select the 10-layer classifier and remove subsequent blocks (ablation) | DeiT-Small · CIFAR-10 | Top-1 accuracy (%): 98.50 -> 98.35 | No clear effect |
| Select the 8-layer classifier and remove subsequent blocks (ablation) | DeiT-Small · CIFAR-10 | Top-1 accuracy (%): 98.50 -> 98.11 | No clear effect |
| Select the 6-layer classifier and remove subsequent blocks (ablation) | DeiT-Small · CIFAR-10 | Top-1 accuracy (%): 98.50 -> 96.55 | Hurt |
| Select the 4-layer classifier and remove subsequent blocks (ablation) | DeiT-Small · CIFAR-10 | Top-1 accuracy (%): 98.50 -> 92.81 | Hurt |
| Select the 2-layer classifier and remove subsequent blocks (ablation) | DeiT-Small · CIFAR-10 | Top-1 accuracy (%): 98.50 -> 74.08 | Hurt |
| Prune 30% of backbone parameters with WDPruning | TransReID with DeiT-Base · Market-1501 | mAP: 88.5% -> 87.7% | Hurt |
Closest papers by meaning
- UPDP: A Unified Progressive Depth Pruner for CNN and Vision Transformer (AAAI 2024 · similarity 0.90)
- Zero-TPrune: Zero-Shot Token Pruning through Leveraging of the Attention Graph in Pre-Trained Transformers (CVPR 2024 · similarity 0.89)
- Collaborative Multi-Mode Pruning for Vision-Language Models (CVPR 2026 · similarity 0.87)
- Data-independent Module-aware Pruning for Hierarchical Vision Transformers (ICLR 2024 · similarity 0.87)
- CORP: Closed-Form One-shot Representation-Preserving Structured Pruning for Transformers (NeurIPS 2026 · similarity 0.86)
- MDP: Multidimensional Vision Model Pruning with Latency Constraint (CVPR 2025 · similarity 0.86)
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.