Research map / Probabilistic and scientific modeling / Causal discovery
Causal Attention for Unbiased Visual Recognition
Tan Wang, Chang Zhou, Qianru Sun, Hanwang Zhang
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 |
|---|---|---|---|
| add the Maxi-Game partition update instead of rebooting training each phase (ablation) | CNN-CaaM · NICO test | accuracy: 44.46 -> 46.62 | Improved |
| add learned partition update instead of randomizing theta each phase (ablation) | CNN-CaaM · NICO test | accuracy: 43.38 -> 46.62 | Improved |
| add learned partition update instead of randomizing theta each phase (ablation) | ViT-CaaM · NICO test | accuracy: 36.15 -> 37.61 | Improved |
| add progressive weight retention instead of rebooting training (ablation) | ViT-CaaM · NICO test | accuracy: 35.46 -> 37.61 | Improved |
| use 2 CaaM blocks instead of 1 (ablation) | CNN-CaaM · NICO test | accuracy: 44.54 -> 45.77 | Improved |
| use 4 CaaM blocks instead of 2 (ablation) | CNN-CaaM · NICO test | accuracy: 45.77 -> 45.31 | No clear effect |
| use m=4 splits instead of m=2 (ablation) | CNN-CaaM · NICO test | accuracy: 44.92 -> 45.77 | No clear effect |
| use m=8 splits instead of m=4 (ablation) | CNN-CaaM · NICO test | accuracy: 45.77 -> 45.74 | No clear effect |
| remove human-annotated partitions for CaaM | CNN-CaaM · ImageNet-A | accuracy: 35.60 -> 38.55 | Improved |
| remove human-annotated partitions for IRM | IRM (CNN) · ImageNet-A | accuracy: 33.52 -> 33.39 | No clear effect |
| add CaaM versus conventional attention | ResNet18+CBAM · NICO test | accuracy: 42.46 -> 46.62 | Improved |
| add CaaM versus conventional attention (attention accuracy) | ResNet18+CBAM · ImageNet-9 attention accuracy | attention accuracy: 69.73 -> 78.37 | Improved |
Closest papers by meaning
- Learning 1D Causal Visual Representation with De-focus Attention Networks (NeurIPS 2024 · similarity 0.83)
- Causal Inference over Visual-Semantic-Aligned Graph for Image Classification (AAAI 2025 · similarity 0.82)
- Towards Causal Foundation Model: on Duality between Optimal Balancing and Attention (ICML 2024 · similarity 0.82)
- Vision-and-Language Navigation via Causal Learning (CVPR 2024 · similarity 0.82)
- From Prompts to Tokens: Internalizing Causal Supervision in Vision-Language Model for Multi-Image Causal Reasoning (ICML 2026 · similarity 0.82)
- Visual Representation Learning through Causal Intervention for Controllable Image Editing (CVPR 2025 · 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.
