Research map / Label-efficient and robust learning / Recognition and long-tail
Caption Bottleneck Models
Seref Baris Cagliyan, Umut Ozdemir, Merve Tapli, Emre Akbas
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 |
|---|---|---|---|
| inject top-ranked class concept into caption (human intervention) (ablation) | CaBM · CUB-200 intervention subset | accuracy (%): 41.7 -> 55.5 (+13.8) | Improved |
| inject top-ranked class concept into caption (ablation) | CaBM · Flowers-102 intervention subset | accuracy (%): 60.6 -> 72.8 | Improved |
| inject top-ranked class concept into caption (ablation) | CaBM · CIFAR-100 intervention subset | accuracy (%): 62.3 -> 75.2 | Improved |
| replace LLM-generated concepts with CaBM-discovered concepts in VLG-CBM (ablation) | VLG-CBM · CUB-200 | ANEC-5 accuracy: 75.79 -> 76.25 | Improved |
| replace LLM-generated concepts with CaBM-discovered concepts in VLG-CBM (ablation) | VLG-CBM · CIFAR-100 | ANEC-5 accuracy: 65.73 -> 66.13 | Improved |
| use text-only caption bottleneck instead of image-domain CBM (ViT) | CaBM vs HybridCBM · CUB-200 | top-1 accuracy: 84.25 -> 76.92 | Hurt |
| use text-only caption bottleneck instead of CNN-backbone CBMs | CaBM vs PS-CBM · ImageNet-1K | top-1 accuracy: 74.00 -> 75.12 | Improved |
| use text-only caption bottleneck | CaBM · Flowers-102 | top-1 accuracy: 97.90 (PS-CBM) -> 86.57 | Hurt |
| use discovered concepts rather than LLM-proposed set (concept quality) | CaBM vs VLG-CBM · CUB-200 | Semantics: 0.86 -> 0.92 | Improved |
Closest papers by meaning
- Hybrid Concept Bottleneck Models (CVPR 2025 · similarity 0.91)
- V2C-CBM: Building Concept Bottlenecks with Vision-to-Concept Tokenizer (AAAI 2025 · similarity 0.90)
- DCBM: Data-Efficient Visual Concept Bottleneck Models (ICML 2025 · similarity 0.90)
- VLG-CBM: Training Concept Bottleneck Models with Vision-Language Guidance (NeurIPS 2024 · similarity 0.90)
- Bayesian Concept Bottleneck Models with LLM Priors (NeurIPS 2025 · similarity 0.89)
- Energy-Based Concept Bottleneck Models: Unifying Prediction, Concept Intervention, and Probabilistic Interpretations (ICLR 2024 · similarity 0.89)
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.
