Research map / Detection and segmentation / CNN features
P2T: Pyramid Pooling Transformer for Scene Understanding
Yu-Huan Wu, Yun Liu, Xin Zhan, Ming‐Ming Cheng
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 four parallel pooling ratios instead of single ratio 8 (ablation) | P2T-Small (ablation) · ImageNet-1K | top-1 accuracy: 73.9 -> 74.7 | Improved |
| use four parallel pooling ratios instead of single ratio 8 (ablation) | P2T-Small (ablation) · ADE20K | mIoU: 34.4 -> 35.7 | Improved |
| use single large pooling ratio 24 instead of 8 (ablation) | P2T-Small (ablation) · ImageNet-1K | top-1 accuracy: 73.9 -> 70.6 | Hurt |
| apply multiple pooling at stage 4 on top of stages 2,3 (ablation) | P2T-Small (ablation) · ImageNet-1K | top-1 accuracy: 74.1 -> 74.5 | Improved |
| apply multiple pooling at all stages (ablation) | P2T-Small (ablation) · ADE20K | mIoU: 34.4 -> 35.7 | Improved |
| use average pooling instead of max pooling (ablation) | P2T-Small (ablation) · ADE20K | mIoU: 33.2 -> 35.7 | Improved |
| use average pooling instead of convolution (ablation) | P2T-Small (ablation) · ImageNet-1K | top-1 accuracy: 73.8 -> 74.7 | Improved |
| use fixed pooling ratios instead of fixed pooled sizes (ablation) | P2T-Small (ablation) · ADE20K | mIoU: 33.3 -> 35.7 | Improved |
| use fixed pooled sizes (saves compute) (ablation) | P2T-Small (ablation) · ImageNet-1K | GFlops: 41.6 -> 38.9 | Mixed |
| add relative positional encoding (ablation) | P2T-Small (ablation) · ADE20K | mIoU: 35.7 -> 37.4 | Improved |
| add IRB after RPE (ablation) | P2T-Small (ablation) · ImageNet-1K | top-1 accuracy: 76.4 -> 79.5 | Improved |
| add overlapping patch embedding after RPE and IRB (ablation) | P2T-Small (ablation) · ADE20K | mIoU: 42.7 -> 44.1 | Improved |
Closest papers by meaning
- Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction Without Convolutions (ICCV 2021 · similarity 0.88)
- MetaFormer Is Actually What You Need for Vision (CVPR 2022 · similarity 0.87)
- Feature Pyramid Transformer (ECCV 2020 · similarity 0.86)
- Rethinking Spatial Dimensions of Vision Transformers (ICCV 2021 · similarity 0.86)
- Pool Me Wisely: On the Effect of Pooling in Transformer-Based Models (NeurIPS 2025 · similarity 0.86)
- SHViT: Single-Head Vision Transformer with Memory Efficient Macro Design (CVPR 2024 · similarity 0.85)
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.
