Research map / Vision-language and multimodal / CLIP and zero-shot
CAPT: Confusion-Aware Prompt Tuning for Reducing Vision-Language Misalignment
Maoyuan Shao, Yutong Gao, Xinyang Huang, Lijuan Sun, Guoshun Nan, Chuang 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 MGDE to fuse semantic and sample confusion information (SEM+SAM -> full) (ablation) | CAPT · Base-to-novel, 11 datasets average | harmonic mean (%): 73.86 -> 83.90 | Improved |
| add sample-level mining on top of semantic-only mining (ablation) | CAPT · Base-to-novel, 11 datasets average | harmonic mean (%): 71.62 -> 73.86 | Improved |
| semantic-level mining alone (ablation) | CAPT · Base-to-novel, 11 datasets average | base accuracy (%): 80.69 vs 87.41 full | Hurt |
| sample-level mining alone (ablation) | CAPT · Base-to-novel, 11 datasets average | novel accuracy (%): 63.41 vs 80.90 full | Hurt |
| add confusion bank statistics to the confusion score (ablation) | CAPT · Base-to-novel, 11 datasets average | harmonic mean (%): 82.62 -> 83.90 | Improved |
| use model's pseudo-GT instead of real ground-truth to select confusion pairs (ablation) | CAPT · Base-to-novel, 11 datasets average | harmonic mean (%): 83.06 -> 83.90 | Improved |
| retrieve representative confusing samples from the bank instead of random samples (ablation) | CAPT · Base-to-novel, 11 datasets average | harmonic mean (%): 83.04 -> 83.90 | Improved |
| use both global attention and local depthwise convolution in the Diff-Manner Adapter (ablation) | CAPT · Grad-CAM qualitative | qualitative attention | Improved |
| switch from PromptKD to CAPT | PromptKD · Base-to-novel, 11 datasets average | harmonic mean (%): 83.73 -> 83.90 | Improved |
| switch from PromptKD to CAPT | PromptKD · Base-to-novel FGVCAircraft | harmonic mean (%): 45.17 -> 44.50 | Hurt |
| switch from TAC to CAPT | TAC · Domain generalization ImageNet-A | accuracy (%): 51.73 -> 57.73 | Improved |
| switch from PromptKD to CAPT | PromptKD · Cross-dataset transfer, average | accuracy (%): 71.33 -> 71.95 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- PromptKD — PromptKD: Unsupervised Prompt Distillation for Vision-Language Models (CVPR 2024) · better on 7 of 8 shared results
- MaPLe — MaPLe: Multi-Modal Prompt Learning (CVPR 2023) · better on 1 of 1 shared result
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
- Unveiling the Tapestry of Consistency in Large Vision-Language Models (NeurIPS 2024 · similarity 0.91)
- Rethinking Misalignment in Vision-Language Model Adaptation from a Causal Perspective (NeurIPS 2024 · similarity 0.90)
- SC-Tune: Unleashing Self-Consistent Referential Comprehension in Large Vision Language Models (CVPR 2024 · similarity 0.89)
- IPO: Interpretable Prompt Optimization for Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- ClawMachine: Learning to Fetch Visual Tokens for Referential Comprehension (ICLR 2025 · similarity 0.89)
- Natural Language Inference Improves Compositionality in Vision-Language Models (ICLR 2025 · 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.
