Research map / Detection and segmentation / CNN features
Conformer: Local Features Coupling Global Representations for Visual Recognition
Zhiliang Peng, Wei Huang, Shanzhi Gu, Lingxi Xie, Yaowei Wang, Jianbin Jiao, 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 |
|---|---|---|---|
| replace DeiT-B with Conformer-B | DeiT-B · ImageNet-1k val | top-1 accuracy: 81.8 -> 84.1 | Improved |
| use Conformer-S/32 backbone instead of ResNet-101 in FPN | ResNet-101 FPN · MSCOCO minival | AP bbox: 39.4 -> 43.1 | Improved |
| use Conformer-S/32 backbone instead of ResNet-101 in Mask R-CNN | ResNet-101 Mask R-CNN · MSCOCO minival | AP segm: 36.1 -> 39.7 | Improved |
| add learnable positional embeddings to DeiT-S (ablation) | DeiT-S · ImageNet-1k val | top-1 accuracy: 77.4 -> 79.8 | Improved |
| add learnable positional embeddings to Conformer-S (ablation) | Conformer-S · ImageNet-1k val | top-1 accuracy: 83.4 -> 83.5 | No clear effect |
| fuse every block instead of every 4 blocks (ablation) | Conformer-S · ImageNet-1k val | top-1 accuracy: 82.2 -> 83.4 | Improved |
| fuse every block instead of every 2 blocks (ablation) | Conformer-S · ImageNet-1k val | top-1 accuracy: 82.9 -> 83.4 | Improved |
| use attention-based sampling instead of avg pooling (ablation) | Conformer-S · ImageNet-1k val | top-1 accuracy: 83.4 -> 83.3 | No clear effect |
| use avg pooling instead of max pooling for sampling (ablation) | Conformer-S · ImageNet-1k val | top-1 accuracy: 83.3 -> 83.4 | No clear effect |
| couple branches with FCU instead of ensembling DeiT-S and ResNet-101 (ablation) | DeiT-S + ResNet-101 · ImageNet-1k val | top-1 accuracy: 81.8 -> 83.4 | Improved |
| use dual structure instead of serial ResNet-50d & DeiT-S hybrid (ablation) | ResNet-50d & DeiT-S · ImageNet-1k val | top-1 accuracy: 80.4 -> 81.9 | Improved |
| increase CNN branch channels from 64 to 256 (ablation) | Conformer (E=384) · ImageNet-1k val | top-1 accuracy: 81.3 -> 83.4 | Improved |
Closest papers by meaning
- Conformer: Local Features Coupling Global Representations for Recognition and Detection (TPAMI 2023 · similarity 0.98)
- TGBFormer: Transformer-GraphFormer Blender Network for Video Object Detection (AAAI 2025 · similarity 0.88)
- UniFormer: Unifying Convolution and Self-Attention for Visual Recognition (TPAMI 2023 · similarity 0.87)
- Conv2Former: A Simple Transformer-Style ConvNet for Visual Recognition (TPAMI 2024 · similarity 0.87)
- Predictive Inference with Feature Conformal Prediction (ICLR 2023 · similarity 0.86)
- Scaling Full Conformal Image Classifiers (NeurIPS 2026 · 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.
