Research map / Label-efficient and robust learning / Noisy labels
Human-Corrected Labels Learning: Enhancing Labels Quality via Human Correction of VLMs Discrepancies
Zhongnian Li, Lan Chen, Yixin Xu, Shi Xu, Xinzheng Xu
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 |
|---|---|---|---|
| Replace Qwen with LLaVA as second VLM (ablation) | HCL · Caltech-101 | accuracy: 94.68 -> 93.33 | Hurt |
| Replace Qwen with LLaVA as second VLM (ablation) | HCL · CIFAR100 | accuracy: 82.48 -> 82.42 | No clear effect |
| Replace Qwen with LLaVA as second VLM (ablation) | HCL · DTD | accuracy: 70.76 -> 69.46 | Hurt |
| Add LLaVA as third VLM to CLIP and Qwen (ablation) | HCL · Caltech-101 | accuracy: 94.68 -> 74.16 | Hurt |
| Add LLaVA as third VLM to CLIP and Qwen (ablation) | HCL · CIFAR100 | accuracy: 82.48 -> 82.42 | No clear effect |
| Add LLaVA as third VLM to CLIP and Qwen (ablation) | HCL · DTD | accuracy: 70.76 -> 70.70 | No clear effect |
| Use simplified English Prompt C instead of Prompt A (ablation) | HCL · Caltech-101 | accuracy: 94.68 -> 94.24 | No clear effect |
| Use simplified English Prompt C instead of Prompt A (ablation) | HCL · DTD | accuracy: 70.76 -> 71.35 | No clear effect |
| Use simplified Prompt C, which lowers VLM label accuracy (ablation) | VLM labels · Caltech-101 | accuracy: 65.38 -> 38.91 | Hurt |
| Increase lambda to mix model predictions into the label distribution (ablation) | HCL · Caltech-101 and EuroSAT | accuracy | Hurt |
| Increase lambda to mix model predictions into the label distribution (ablation) | HCL · DTD | accuracy | Improved |
| Add human correction on VLM-discrepant samples to CLIP labels | CLIP labels · EuroSAT annotation accuracy | accuracy: 43.21 -> 95.20 | Improved |
Closest papers by meaning
- Candidate Pseudolabel Learning: Enhancing Vision-Language Models by Prompt Tuning with Unlabeled Data (ICML 2024 · similarity 0.87)
- Revisiting Interpolation for Noisy Label Correction (AAAI 2025 · similarity 0.87)
- Adaptive Integration of Partial Label Learning and Negative Learning for Enhanced Noisy Label Learning (AAAI 2024 · similarity 0.86)
- Handling Imbalanced Pseudolabels for Vision-Language Models with Concept Alignment and Confusion-Aware Calibrated Margin (ICML 2025 · similarity 0.85)
- Mitigating Label Noise through Data Ambiguation (AAAI 2024 · similarity 0.85)
- ABC-Align: Prediction-Powered Alignment with Adaptive Bias Control (NeurIPS 2026 · similarity 0.85)
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.
