Research map / Detection and segmentation / CNN features
Fast-iTPN: Integrally Pre-Trained Transformer Pyramid Network With Token Migration
Yunjie Tian, Lingxi Xie, Jihao Qiu, Jianbin Jiao, Yaowei Wang, Qi Chuan Tian, Qixiang Ye
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 |
|---|---|---|---|
| add integral pre-training of backbone and feature pyramid (ablation) | iTPN (400 ep) · ImageNet-1K | top-1 accuracy: 84.4 -> 85.1 | Improved |
| add integral pre-training (ablation) | iTPN (400 ep) · COCO box AP | box AP: 50.6 -> 51.5 | Improved |
| load the pre-trained feature pyramid for detection (ablation) | iTPN · COCO box AP | box AP: 51.5 -> 52.1 | Improved |
| load the pre-trained feature pyramid for segmentation (ablation) | iTPN · ADE20K | mIoU: 51.8 -> 52.2 | Improved |
| use C-MLP instead of convolution in backbone and feature pyramid (ablation) | iTPN · ImageNet-1K | top-1 accuracy: 84.3 -> 84.9 | Improved |
| add masked feature modeling (MFM) (ablation) | iTPN with C-MLP · ImageNet-1K | top-1 accuracy: 84.9 -> 85.1 | Improved |
| add masked feature modeling (MFM) (ablation) | iTPN with C-MLP · COCO | box AP: 51.8 -> 52.1 | Improved |
| increase token migration ratio from 10% to 40% (ablation) | Fast-iTPN-B · COCO | mAP: 53.4 -> 53.0 | Mixed |
| replace MLP with ConvMLP (ablation) | Fast-iTPN · COCO | speed: 10% speed improvement | Improved |
| apply window attention (ablation) | Fast-iTPN-B · DOTA | mAP: 79.3 -> 77.7 | Hurt |
| add gathering tokens to window attention (ablation) | Fast-iTPN-B · DOTA | mAP: 77.7 -> 78.6 | Improved |
| add relative position encoding for window attention (HRPE) (ablation) | Fast-iTPN-B · DOTA | mAP: 77.7 -> 79.0 | Improved |
Closest papers by meaning
- Integrally Pre-Trained Transformer Pyramid Networks (CVPR 2023 · similarity 0.94)
- Feature Pyramid Networks for Object Detection (CVPR 2017 · similarity 0.87)
- Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction Without Convolutions (ICCV 2021 · similarity 0.87)
- NAS-FPN: Learning Scalable Feature Pyramid Architecture for Object Detection (CVPR 2019 · similarity 0.86)
- FaPN: Feature-Aligned Pyramid Network for Dense Image Prediction (ICCV 2021 · similarity 0.86)
- FasterViT: Fast Vision Transformers with Hierarchical Attention (ICLR 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.
