Research map / Detection and segmentation / CNN features
Dynamic Grained Encoder for Vision Transformers
Lin Song, Songyang Zhang, Songtao Liu, Zeming Li, Xuming He, Hongbin Sun, Jian Sun, Nanning Zheng
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 |
|---|---|---|---|
| use region-wise routing instead of layer-wise routing (ablation) | PVT-S+DGE · ImageNet val | Top-1 accuracy: 79.1 -> 80.2 | Improved |
| restrict granularity set to binary mask {0,1} instead of {1,2,4} (ablation) | PVT-S+DGE · ImageNet val | Top-1 accuracy: 80.2 -> 78.8 | Hurt |
| use granularity set {1,2} instead of {1,2,4} (ablation) | PVT-S+DGE · ImageNet val | Top-1 accuracy: 80.2 -> 80.0 | No clear effect |
| add granularity 8 to candidate set {1,2,4} (ablation) | PVT-S+DGE · ImageNet val | Top-1 accuracy: 80.2 -> 79.9 | No clear effect |
| add DGE to PVT-S at budget 0.5 (layer-wise variant) | PVT-S · ImageNet val | FLOPs: 6.2G -> 3.4G | Improved |
| add DGE (region-wise, Phi={1,2,4}) to PVT-S | PVT-S · ImageNet val | Top-1 accuracy: 80.2 -> 80.2 | No clear effect |
| add DGE to PVT-S backbone | PVT-S · COCO val Mask R-CNN 1x | backbone FLOPs: 251 -> 185 | Improved |
| add DGE to PVT-S backbone | PVT-S · COCO val Mask R-CNN 1x | box AP: 40.4 -> 40.1 | No clear effect |
| add DGE to PVT-S backbone | PVT-S · COCO val backbone latency CPU | latency (ms): 880 -> 440 | Improved |
| add DGE to DPVT-M backbone | DPVT-M · COCO val Mask R-CNN 1x | box AP: 46.4 -> 45.8 | Hurt |
| add DGE to DPVT-M backbone | DPVT-M · COCO val Mask R-CNN 1x | backbone FLOPs: 236 -> 169 | Improved |
| add DGE to DPVT-M | DPVT-M · ADE-20K val Semantic-FPN | mIoU: 46.8 -> 46.1 | Hurt |
Closest papers by meaning
- Benchmarking Large Vision-Language Models on Fine-Grained Image Tasks: A Comprehensive Evaluation (ICLR 2026 · similarity 0.88)
- Scaling White-Box Transformers for Vision (NeurIPS 2024 · similarity 0.87)
- GranViT: A Fine-Grained Vision Model With Autoregressive Perception For MLLMs (ICLR 2026 · similarity 0.86)
- ImagePiece: Content-aware Re-tokenization for Efficient Image Recognition (AAAI 2025 · similarity 0.86)
- Frozen Transformers in Language Models Are Effective Visual Encoder Layers (ICLR 2024 · similarity 0.86)
- Dynamic Convolutions Improve Transformers (NeurIPS 2026 · 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.
