Research map / Detection and segmentation / CNN features
Dynamic Token Pruning in Plain Vision Transformers for Semantic Segmentation
Quan Tang, Bowen Zhang, Jiajun Liu, Fagui Liu, Yifan Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| finetune heads for 40k iterations with DToP instead of applying directly (ablation) | DToP · ADE20K | mIoU (%): 47.9 -> 49.8 | Improved |
| retrain the entire model with DToP instead of finetuning heads (ablation) | DToP · ADE20K | mIoU (%): 49.8 -> 49.1 | Hurt |
| keep top-k tokens per category when pruning (ablation) | DToP · ADE20K | mIoU (%): 48.7 -> 49.5 | Improved |
| combine average and top-k context tokens (ablation) | DToP · ADE20K | mIoU (%): 49.5 -> 49.7 | Improved |
| lower confidence threshold p0 from 0.95 to 0.60 (ablation) | DToP with ATM head · ADE20K | mIoU (%): 49.8 -> 46.8 | Hurt |
| split backbone into 4 stages instead of 3 (ablation) | DToP · ADE20K | mIoU (%): 49.8 -> 48.3 | Hurt |
| use a powerful ATM auxiliary head instead of FCN for grading difficulty (ablation) | DToP · ADE20K | mIoU (%): 46.9 -> 48.4 | Improved |
| apply DToP@Finetune to SETR | SETR ViT-Base · ADE20K | GFLOPs: 107.7 -> 80.6 | Improved |
| apply DToP@Finetune to SegViT ViT-Large | SegViT · ADE20K | mIoU (%): 53.3 -> 52.4 with 617.0 -> 380.3 GFLOPs | Mixed |
| apply DToP to Pascal Context | SETR ViT-Base · Pascal Context | mIoU (%): 58.1 -> 58.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- SegViT — SegViT: Semantic Segmentation with Plain Vision Transformers (NeurIPS 2022) · better on 1 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Token Cropr: Faster ViTs for Quite a Few Tasks (CVPR 2025 · similarity 0.90)
- ALGM: Adaptive Local-then-Global Token Merging for Efficient Semantic Segmentation with Plain Vision Transformers (CVPR 2024 · similarity 0.89)
- DTMFormer: Dynamic Token Merging for Boosting Transformer-Based Medical Image Segmentation (AAAI 2024 · similarity 0.89)
- ImagePiece: Content-aware Re-tokenization for Efficient Image Recognition (AAAI 2025 · similarity 0.89)
- Balanced Token Pruning: Accelerating Vision Language Models Beyond Local Optimization (NeurIPS 2025 · similarity 0.89)
- Structural Pruning via Spatial-aware Information Redundancy for Semantic Segmentation (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.
