Research map / Large language models / Instruction and fine-tuning
Vocabulary Scaling Law: Tuning Open-vocabulary Predictors for Their Openness
Ziliang Chen, Yulu Li, Liangda Fang, Jusheng Zhang, Yongsen Zheng, Quanlong Guan, Xipeng Chen
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 |
|---|---|---|---|
| tune class-name embeddings instead of context prompt (vocabulary U) (ablation) | CLIP ViT-B/16 prompt tuning · CIFAR100 | Extensibility gap vs closed-set: -10.9 -> -4.2 | Improved |
| tune class names on full vocabulary U instead of target vocabulary only (ablation) | Name-based PT · CIFAR100 | Acc-E: 79.4 -> 81.4 | Improved |
| add orthogonality constraint between target and open-set class embeddings (ablation) | Name-based PT (U) · CIFAR100 | Acc-E: 81.4 -> 83.7 | Improved |
| use full vocabulary U for context-based prompt tuning (ablation) | Context-based PT · CIFAR100 | Acc-E: 76.9 -> 73.2 | Hurt |
| scale up CLIP backbone (neural scaling) | CLIP zero-shot · CIFAR100, Entity13, Living17 | stability/extensibility gaps | No clear effect |
| replace linear greedy submodular selection with Random, Bi-Search or Full (ablation) | SVFT · Birds, Rare Species | stability and extensibility | Hurt |
| add class names of U/V(T) during fine-tuning of baselines | MAPLE, CLIP-Adapter · Birds, Rare Species | accuracy | Improved |
Closest papers by meaning
- Transferable and Principled Efficiency for Open-Vocabulary Segmentation (CVPR 2024 · similarity 0.88)
- Renovating Names in Open-Vocabulary Segmentation Benchmarks (NeurIPS 2024 · similarity 0.87)
- Understanding Fine-tuning CLIP for Open-vocabulary Semantic Segmentation in Hyperbolic Space (CVPR 2025 · similarity 0.87)
- Scaling Laws for Robust Comparison of Open Foundation Language-Vision Models and Datasets (NeurIPS 2025 · similarity 0.87)
- CRoFT: Robust Fine-Tuning with Concurrent Optimization for OOD Generalization and Open-Set OOD Detection (ICML 2024 · similarity 0.87)
- Open-Vocabulary Calibration for Fine-tuned CLIP (ICML 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.
