Research map / Architectures and efficiency / Attention and transformers
Active Token Mixer
Guoqiang Wei, Zhizheng Zhang, Cuiling Lan, Yan Lu, Zhibo Chen
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 ATMNet instead of scale-comparable backbones | ImageNet-1K classification · ImageNet-1K val, 224² | Top-1 accuracy (%): 82.0 vs 81.7 | Improved |
| Use ATMNet instead of Twins-L | Semantic FPN · ADE20K val | mIoU: 48.1 vs 46.7 | Improved |
| Use ATMNet-L instead of Twins-L | UperNet · ADE20K val, multi-scale | mIoU: 51.1 vs 50.2 | Improved |
| Use ATMNet-T instead of WaveMLP-S | RetinaNet 1× · COCO val2017 | AP^b: 43.6 vs 43.4 | Improved |
| Use ATMNet-L instead of Twins-L | Mask R-CNN 1× · COCO val2017 | AP^b: 47.4 vs 45.9 | Improved |
| Add adaptive spatial token mixing rather than fixing all offsets to 0 (ablation) | ATMNet · ImageNet-1K val | Top-1 accuracy (%): 79.3 -> 82.0 | Improved |
| Add adaptive spatial token mixing rather than fixing all offsets to 0 (ablation) | ATMNet · COCO, RetinaNet 1× | AP^b: 36.0 -> 43.6 | Improved |
| Add adaptive spatial token mixing rather than fixing all offsets to 0 (ablation) | ATMNet · ADE20K, Semantic FPN | mIoU: 37.9 -> 45.8 | Improved |
| Add PEG for offset-generation position information (ablation) | ATMNet · ImageNet-1K val | Top-1 accuracy (%): 82.0 -> 82.0 | No clear effect |
| Add PEG for offset-generation position information (ablation) | ATMNet · COCO, RetinaNet 1× | AP^b: 43.4 -> 43.6 | Improved |
| Add PEG for offset-generation position information (ablation) | ATMNet · ADE20K, Semantic FPN | mIoU: 45.6 -> 45.8 | Improved |
| Replace FPN with ATMFPN (ablation) | ResNet-50 · ADE20K, Semantic FPN | mIoU: 37.3 -> 40.3 (+3.0) | Improved |
Closest papers by meaning
- CanViT: Toward Active-Vision Foundation Models (NeurIPS 2026 · similarity 0.83)
- FFT-Based Dynamic Token Mixer for Vision (AAAI 2024 · similarity 0.83)
- Pixels to Tokens: Token Space Efficient Active Learning for Low-Budget Semantic Segmentation (NeurIPS 2026 · similarity 0.83)
- FreeNet: Liberating Depth-Wise Separable Operations for Building Faster Mobile Vision Architectures (AAAI 2025 · similarity 0.82)
- PredToken: Predicting Unknown Tokens and Beyond with Coarse-to-Fine Iterative Decoding (CVPR 2024 · similarity 0.82)
- Hydra: Bidirectional State Space Models Through Generalized Matrix Mixers (NeurIPS 2024 · similarity 0.82)
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.
