Research map / Architectures and efficiency / Attention and transformers
Inception Transformer
Chenyang Si, Weihao Yu, Pan Zhou, Yichen Zhou, Xinchao Wang, Shuicheng YAN
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 max-pooling high-frequency branch to attention-only mixer (ablation) | iFormer-S · ImageNet-1K 100-epoch ablation | top-1 accuracy: 80.8 -> 81.0 | Improved |
| add depthwise conv branch on top of attention+MaxPool (ablation) | iFormer-S · ImageNet-1K 100-epoch ablation | top-1 accuracy: 81.0 -> 81.2 | Improved |
| replace attention-only mixer with Inception mixer (ablation) | iFormer-S · ImageNet-1K 100-epoch ablation | FLOPs (G): 5.2 -> 4.8 | Improved |
| use frequency ramp (fewer high-frequency channels in deeper layers) instead of other channel-ratio schedules (ablation) | iFormer-S · ImageNet-1K 100-epoch ablation | top-1 accuracy: 80.5 -> 81.2 | Improved |
| iFormer-S vs DeiT-S | iFormer-S · ImageNet-1K val | top-1 accuracy: 79.8 -> 83.4 | Improved |
| iFormer-S backbone vs ResNet50 in Mask R-CNN | iFormer-S · COCO val2017 | box AP: 38.0 -> 46.2 | Improved |
| iFormer-S vs UniFormer-S on ADE20K Semantic FPN | iFormer-S · ADE20K | mIoU: 46.6 -> 48.6 | Improved |
Closest papers by meaning
- Frequency-Dynamic Attention Modulation For Dense Prediction (ICCV 2025 · similarity 0.86)
- Improving Vision Transformers by Revisiting High-Frequency Components (ECCV 2022 · similarity 0.86)
- Anti-Oversmoothing in Deep Vision Transformers via the Fourier Domain Analysis: From Theory to Practice (ICLR 2022 · similarity 0.85)
- IFORMER: INTEGRATING CONVNET AND TRANSFORMER FOR MOBILE APPLICATION (ICLR 2025 · similarity 0.85)
- Feature Modulation Transformer: Cross-Refinement of Global Representation via High-Frequency Prior for Image Super-Resolution (ICCV 2023 · similarity 0.85)
- Exploring Frequency-Inspired Optimization in Transformer for Efficient Single Image Super-Resolution (TPAMI 2025 · 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.
