Research map / Architectures and efficiency / Attention and transformers
Expediting Large-Scale Vision Transformer for Dense Prediction without Fine-tuning
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 |
|---|---|---|---|
| replace token clustering with adaptive average pooling (ablation) | Ours on Segmenter+ViT-L/16 · ADE20K (28x28) | mIoU: 46.54 -> 51.56 | Helped |
| replace token reconstruction with bilinear upsample (ablation) | Ours on Segmenter+ViT-L/16 · ADE20K (28x28) | mIoU: 44.74 -> 51.56 | Helped |
| reduce cluster size from 28x28 to 8x8 (ablation) | Ours on Segmenter+ViT-L/16 · ADE20K | mIoU: 51.56 -> 32.13 | Hurt |
| reduce cluster size from 40x40 baseline to 28x28 (ablation) | Segmenter+ViT-L/16 · ADE20K | mIoU: 51.82 -> 51.56; GFLOPs 659.0 -> 438.9 | Mixed |
| vary number of neighbors lambda (3x3 to 7x7) (ablation) | token clustering · ADE20K | mIoU: 51.46 / 51.56 / 51.55 | No clear effect |
| vary temperature tau (ablation) | token clustering · ADE20K | mIoU: 51.18 -> 51.56 -> 51.07 (tau 10, 50, 75) | Mixed |
| vary k in k-NN (ablation) | token clustering · ADE20K | mIoU: 51.34 / 51.56 / 51.56 | No clear effect |
| apply method to lighter ViT-B/16 | Segmenter+ViT-B/16 · ADE20K | mIoU: 48.48 -> 48.40; GFLOPs 124.7 -> 91.9 | No clear effect |
| apply method to DPT | DPT R50+ViT-B/16 · NYUv2 | AbsRel: 0.110 -> 0.113 | Hurt |
Closest papers by meaning
- Win-Win: Training High-Resolution Vision Transformers from Two Windows (ICLR 2024 · similarity 0.88)
- Token Cropr: Faster ViTs for Quite a Few Tasks (CVPR 2025 · similarity 0.87)
- Memory Efficient Transformer Adapter for Dense Predictions (ICLR 2025 · similarity 0.87)
- Resource-Efficient Transformer Pruning for Finetuning of Large Models (CVPR 2024 · similarity 0.86)
- Accelerating Transformers with Spectrum-Preserving Token Merging (NeurIPS 2024 · similarity 0.86)
- An Image is Worth More Than 16x16 Patches: Exploring Transformers on Individual Pixels (ICLR 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.