Research map / Detection and segmentation / CNN features
EATFormer: Improving Vision Transformer Inspired by Evolutionary Algorithm
Jiangning Zhang, Xiangtai Li, Yabiao Wang, Chengjie Wang, Yibo Yang, Yong Liu, Dacheng Tao
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 EATFormer-Mobile instead of MobileNetV3-Small 0.75x | mobile ViT · ImageNet-1K val (224x224, from scratch) | top-1 accuracy: 65.4 -> 69.4 | Improved |
| raise input resolution from 224 to 384 for Tiny | EATFormer-Tiny · ImageNet-1K val | top-1 accuracy: 78.4 -> 80.1 | Improved |
| raise input resolution from 224 to 384 for Small | EATFormer-Small · ImageNet-1K val | top-1 accuracy: 83.1 -> 84.3 | Improved |
| raise input resolution from 224 to 384 for Base | EATFormer-Base · ImageNet-1K val | top-1 accuracy: 83.9 -> 84.9 | Improved |
| use improved EATFormer pyramid design over previous conference columnar EAT (Tiny) | EATFormer-Tiny · ImageNet-1K val | top-1 accuracy: +5.7 | Improved |
| use improved EATFormer pyramid design over previous conference columnar EAT (Small) | EATFormer-Small · ImageNet-1K val | top-1 accuracy: +2.7 | Improved |
| use improved EATFormer pyramid design over previous conference columnar EAT (Base) | EATFormer-Base · ImageNet-1K val | top-1 accuracy: +1.9 | Improved |
| use EATFormer-Base instead of MPViT-B | backbone · ImageNet-1K val | GPU throughput (images/s): 181 -> 329 | Improved |
| use EATFormer-Tiny instead of PVT-Tiny backbone | Mask R-CNN · COCO (1x/3x) | box AP: +5.6/+5.6 | Improved |
| use EATFormer-Tiny instead of MPViT-T backbone | Mask R-CNN · COCO (3x schedule) | box AP: 44.8 -> 45.4 | Improved |
| use EATFormer-Small instead of MPViT-S backbone | Mask R-CNN · COCO (1x schedule) | box AP: 46.4 -> 46.1 | Hurt |
| use EATFormer-Tiny instead of VAN-Tiny backbone | UperNet · ADE20K val | mIoU: 41.1 -> 44.5 | Improved |
Closest papers by meaning
- Analogous to Evolutionary Algorithm: Designing a Unified Sequence Model (NeurIPS 2021 · similarity 0.91)
- AdaViT: Adaptive Vision Transformers for Efficient Image Recognition (CVPR 2022 · similarity 0.87)
- AdaptFormer: Adapting Vision Transformers for Scalable Visual Recognition (NeurIPS 2022 · similarity 0.86)
- AutoFormer: Searching Transformers for Visual Recognition (ICCV 2021 · similarity 0.86)
- Auto-Prox: Training-Free Vision Transformer Architecture Search via Automatic Proxy Discovery (AAAI 2024 · similarity 0.85)
- Memory Efficient Transformer Adapter for Dense Predictions (ICLR 2025 · similarity 0.84)
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.
