Research map / Architectures and efficiency / Attention and transformers
ViT-NeT: Interpretable Vision Transformers with Neural Tree Decoder
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 |
|---|---|---|---|
| add contextual transformer module (CTM) to tree edges (ablation) | DeiT-B backbone with NeT · CUB-200-2011 | Top-1 accuracy (%): +0.8 | Helped |
| add contextual transformer module (CTM) to tree edges (ablation) | SwinT-B backbone with NeT · CUB-200-2011 | Top-1 accuracy (%): +0.9 | Helped |
| use global max pooling (GMP) instead of global average pooling (GAP) in leaf prediction (ablation) | ViT-NeT leaf nodes (SwinT-B) · CUB-200-2011 | Top-1 accuracy (%): +0.2 | Helped |
| tree depth 4 versus 3, 5+ (depth sweep) (ablation) | ViT-NeT tree decoder · CUB-200-2011 | Top-1 accuracy (%) | Mixed |
| increase tree depth beyond 6 (depth sweep 5-7) (ablation) | ViT-NeT tree decoder · Stanford Cars and Stanford Dogs | Top-1 accuracy (%) | Mixed |
| add NeT decoder on top of SwinT-B backbone | SwinT-B · CUB-200-2011 | Top-1 accuracy (%): +3.2 | Helped |
| add NeT decoder on top of DeiT-B backbone | DeiT-B · CUB-200-2011 | Top-1 accuracy (%): +2.5 | Helped |
| add NeT decoder on top of DeiT-B backbone | DeiT-B · Stanford Dogs | Top-1 accuracy (%): +2.1 | Helped |
| add NeT decoder on top of SwinT-B backbone | SwinT-B · Stanford Dogs | Top-1 accuracy (%): +2.3 | Helped |
| add NeT decoder on top of DeiT-B backbone | DeiT-B · Stanford Cars | Top-1 accuracy (%): +2.3 | Helped |
| add NeT decoder on top of SwinT-B backbone | SwinT-B · Stanford Cars | Top-1 accuracy (%): +0.5 | Helped |
Closest papers by meaning
- ViTree: Single-Path Neural Tree for Step-Wise Interpretable Fine-Grained Visual Categorization (AAAI 2024 · similarity 0.92)
- Intriguing Properties of Vision Transformers (NeurIPS 2021 · similarity 0.89)
- Attention Guided CAM: Visual Explanations of Vision Transformer Guided by Self-Attention (AAAI 2024 · similarity 0.88)
- Vision Transformers Need Registers (ICLR 2024 · similarity 0.87)
- Decomposing and Interpreting Image Representations via Text in ViTs Beyond CLIP (NeurIPS 2024 · similarity 0.87)
- Bi-ViT: Pushing the Limit of Vision Transformer Quantization (AAAI 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.