Research map / Vision-language and multimodal / CLIP and zero-shot
Revisiting Confidence Calibration for Misclassification Detection in VLMs
Jincheng Huang, Jie Xu, Xiaoshuang Shi, Ping Hu, Lei Feng, Xiaofeng Zhu
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 LMN calibration | CLIP · DTD | AUROC: 0.762 -> 0.802 | Improved |
| add LMN calibration | CLIP · EuroSAT | AUROC: 0.65 -> 0.788 | Improved |
| add LMN calibration | CLIP · MNIST | AUROC: 0.813 -> 0.915 | Improved |
| add LMN calibration | CLIP · CUB | AUROC: 0.807 -> 0.812 | Improved |
| add full-probability constraint only (ablation) | CLIP · EuroSAT | AUROC: 0.650 -> 0.778 | Improved |
| add full-probability constraint only (ablation) | CLIP · MNIST | FPR90-S: 0.511 -> 0.228 | Improved |
| add the remaining LMN components on top of FPC (ablation) | CLIP · EuroSAT | AUROC: 0.778 -> 0.788 | Improved |
| add the remaining LMN components on top of FPC (ablation) | CLIP · CUB | AUROC: 0.808 -> 0.812 | No clear effect |
| use ViLU instead of zero-shot CLIP | CLIP · CUB | AUROC: 0.807 -> 0.801 | Hurt |
| use FeatureClipping instead of zero-shot CLIP | CLIP · DTD | AUROC: 0.762 -> 0.749 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ViLU — ViLU: Learning Vision-Language Uncertainties for Failure Prediction (ICCV 2025) · better on 6 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- An Empirical Study Into What Matters for Calibrating Vision-Language Models (ICML 2024 · similarity 0.89)
- Understanding and Mitigating Miscalibration in Prompt Tuning for Vision-Language Models (ICML 2025 · similarity 0.87)
- Open-Vocabulary Calibration for Fine-tuned CLIP (ICML 2024 · similarity 0.87)
- Linking Perception, Confidence and Accuracy in MLLMs (CVPR 2026 · similarity 0.87)
- Beyond Accuracy: On the Effects of Fine-Tuning Towards Vision-Language Model’s Prediction Rationality (AAAI 2025 · similarity 0.87)
- VLM-RobustBench: A Comprehensive Benchmark for Robustness of Vision-Language Models (ICML 2026 · 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.
