Research map / Architectures and efficiency / Neural architecture search
"UniNet: Unified Architecture Search with Convolution, Transformer, and MLP"
Jihao Liu, Xin Huang, Guanglu Song, Hongsheng Li, Yu Liu
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 |
|---|---|---|---|
| search over convolution, transformer and MLP operators (vs convolution only) (ablation) | UniNet-B0 · ImageNet-1K val (ablation) | top-1 accuracy (%): 77.7 -> 79.1 | Improved |
| search DSM per stage (vs L-DSM in all stages) (ablation) | UniNet-B0 · ImageNet-1K val (ablation) | top-1 accuracy (%): 78.5 -> 79.1 | Improved |
| search DSM per stage (vs G-DSM in all stages) (ablation) | UniNet-B0 · ImageNet-1K val (ablation) | top-1 accuracy (%): 76.8 -> 79.1 | Improved |
| search DSM per stage (vs LG-DSM in all stages) (ablation) | UniNet-B0 · ImageNet-1K val (ablation) | top-1 accuracy (%): 78.9 -> 79.1 | Improved |
| add context-aware DSMs to PVT-Tiny (ablation) | PVT-Tiny · ImageNet-1K val | top-1 accuracy (%): 75.1 -> 77.5 | Improved |
| add context-aware DSMs to Swin-T (ablation) | Swin-T · ImageNet-1K val | top-1 accuracy (%): 81.2 -> 81.6 | Improved |
| stack MLP-Mixer blocks before ViT blocks (MLP-ViT) instead of pure ViT (ablation) | ViT · ImageNet-1K val (pilot, operator stacking) | top-1 accuracy (%): 78.0 -> 77.8 | No clear effect |
| stack ViT blocks before MLP blocks (ViT-MLP) instead of pure ViT (ablation) | ViT · ImageNet-1K val (pilot, operator stacking) | top-1 accuracy (%): 78.0 -> 76.5 | Hurt |
| use pure MLP-Mixer instead of ViT (ablation) | ViT · ImageNet-1K val (pilot, operator stacking) | top-1 accuracy (%): 78.0 -> 76.8 | Hurt |
Closest papers by meaning
- AutoFormer: Searching Transformers for Visual Recognition (ICCV 2021 · similarity 0.88)
- Ubernet: Training a Universal Convolutional Neural Network for Low-, Mid-, and High-Level Vision Using Diverse Datasets and Limited Memory (CVPR 2017 · similarity 0.88)
- Win-Win: Training High-Resolution Vision Transformers from Two Windows (ICLR 2024 · similarity 0.87)
- Training-Free Transformer Architecture Search (CVPR 2022 · similarity 0.87)
- Vision-Oriented Lightweight Neural Architecture Search with Budget-Adaptive Evaluation (CVPR 2026 · similarity 0.87)
- Vision Transformers: From Semantic Segmentation to Dense Prediction (IJCV 2024 · similarity 0.87)
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.
