Research map / Detection and segmentation / CNN features
Global Context Vision Transformers
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 local-only attention with global context self-attention (ablation) | GC ViT-T · ImageNet-1K | Top-1 accuracy: 82.6 -> 83.5 | Helped |
| replace local-only attention with global context self-attention (ablation) | GC ViT-T · ADE20K | mIoU: 45.8 -> 47.0 | Helped |
| remove window shifting (ablation) | Swin-T · ImageNet-1K | Top-1 accuracy: 81.3 -> 80.2 | Hurt |
| reparameterize window size, block count, and MLP ratio (ablation) | Swin-T · ImageNet-1K | Top-1 accuracy: +1.7 | Helped |
| add the GC ViT CNN-based stem (ablation) | reparameterized Swin-T · ImageNet-1K | Top-1 accuracy: +0.3 | Helped |
| add the modified Fused-MBConv downsampler (ablation) | GC ViT-T stem variant · ImageNet-1K | Top-1 accuracy: +0.4 | Helped |
| replace modified Fused-MBConv downsampling with convolution and max pooling (ablation) | GC ViT-T · ImageNet-1K | Top-1 accuracy: 83.5 -> 82.7 | Hurt |
| replace modified Fused-MBConv downsampling with Swin-style linear patch merging (ablation) | GC ViT-T · ImageNet-1K | Top-1 accuracy: 83.5 -> 82.9 | Hurt |
| add the GC module without window shifting (ablation) | Swin-T · ImageNet-1K | Top-1 accuracy: +0.9% | Helped |
| replace global query with global keys and values (ablation) | GC ViT-T · ADE20K | mIoU: 47.0 -> 44.6 | Hurt |
| replace global query with global values (ablation) | GC ViT-T · ADE20K | mIoU: 47.0 -> 45.1 | Hurt |
| increase local/global batch sizes from 32/1024 to 128/4096 (ablation) | GC ViT-T without EMA · ImageNet-1K | Top-1 accuracy: 83.45 -> 83.46 | No clear effect |
Closest papers by meaning
- LF-ViT: Reducing Spatial Redundancy in Vision Transformer for Efficient Image Recognition (AAAI 2024 · similarity 0.87)
- xT: Nested Tokenization for Larger Context in Large Images (ICML 2024 · similarity 0.86)
- You Only Need Less Attention at Each Stage in Vision Transformers (CVPR 2024 · similarity 0.86)
- Your ViT is Secretly an Image Segmentation Model (CVPR 2025 · similarity 0.86)
- Attention Guided CAM: Visual Explanations of Vision Transformer Guided by Self-Attention (AAAI 2024 · similarity 0.86)
- FasterViT: Fast Vision Transformers with Hierarchical Attention (ICLR 2024 · 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.