Research map / Vision-language and multimodal / CLIP and zero-shot
Conceptual Codebook Learning for Vision-Language Models
Yi Zhang, Ke Yu, Siqi Wu, Zhihai He
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 learnable conceptual codebook with handcrafted concept cache regularization over CPL | CLIP ViT-B/16 prompt tuning · Base-to-novel (11-dataset average) | harmonic mean accuracy: 81.08 -> 82.70 | Improved |
| remove orthogonality loss L_or (ablation) | CoCoLe · Base-to-novel (ImageNet) | harmonic mean accuracy: -0.46 when removed from 76.84 | Improved |
| add consistency regularization L_cc to handcrafted concept prompts (ablation) | CoCoLe · Base-to-novel (ImageNet) | harmonic mean accuracy: -1.98 when removed from 76.84 | Improved |
| add matching loss L_ma to learn codebook keys (ablation) | CoCoLe · Base-to-novel (ImageNet) | harmonic mean accuracy: -2.39 when removed from 76.84 | Improved |
| select too many concepts K3 simultaneously during training | CoCoLe · Base-to-novel (ImageNet) | harmonic mean accuracy | Hurt |
| use overly long prompts | CoCoLe · Base-to-novel (ImageNet) | harmonic mean accuracy | Hurt |
| CoCoLe versus CPL on base classes of Caltech101, Flowers102, DTD | CoCoLe · Base-to-novel (Caltech101) | base accuracy: 98.35 -> 98.17 | Mixed |
| CoCoLe versus CPL on training time | CoCoLe · Training time (11-dataset average) | training time per image: 25ms -> 10ms | Improved |
| CoCoLe versus CPL on cross-dataset transfer | CoCoLe · Cross-dataset (UCF101 target) | accuracy: 70.52 -> 72.78 | Improved |
Closest papers by meaning
- Concept-Guided Prompt Learning for Generalization in Vision-Language Models (AAAI 2024 · similarity 0.90)
- Synthesize Diagnose and Optimize: Towards Fine-Grained Vision-Language Understanding (CVPR 2024 · similarity 0.89)
- Bridging the gap to real-world language-grounded visual concept learning (NeurIPS 2025 · similarity 0.89)
- Concept-Aware Batch Sampling Improves Language-Image Pretraining (CVPR 2026 · similarity 0.88)
- TextRefiner: Internal Visual Feature as Efficient Refiner for Vision-Language Models Prompt Tuning (AAAI 2025 · similarity 0.88)
- Language-Informed Visual Concept Learning (ICLR 2024 · similarity 0.88)
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.
