Research map / Vision-language and multimodal / CLIP and zero-shot
Rethinking BCE Loss for Multi-Label Image Recognition with Fine-Tuning
Ao Zhou, Zhiwei Jiang, Zifeng Cheng, Cong Wang, Yafeng Yin, Shufan Yang, Qing Gu
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 class-wise covariance regularization (CCR) to BCE fine-tuning | CoOp · Six datasets avg | ECE (x10^-2): 13.25 -> 7.35 | Improved |
| add CCR to BCE fine-tuning | CoOp · New classes (six datasets avg) | ECE (x10^-2): 21.58 -> 11.03 | Improved |
| add CCR to BCE fine-tuning | T2I-PAL · Six datasets avg | ECE (x10^-2): 4.09 -> 3.62 | Improved |
| add CCR to DualCoOp | DualCoOp · Six datasets avg | ECE (x10^-2): 4.53 -> 4.57 | No clear effect |
| add CCR to TaI-DPT | TaI-DPT · COCO-LT target (domain generalization) | ECE (x10^-2): 5.23 -> 5.58 | Hurt |
| add CCR to CoOp | CoOp · MS-COCO source accuracy | accuracy: 69.44 -> 71.47 | Improved |
| add CCR to TaI-DPT | TaI-DPT · COCO-2014 target accuracy | accuracy: 69.57 -> 74.94 | Improved |
| apply DAC or DOR calibration instead of CCR to CoOp | CoOp · Tail classes | ECE (x10^-2): 8.37 -> 9.28 (DAC) | Mixed |
| vary regularization coefficient lambda (ablation) | CCR · base classes, CoOp | accuracy and ECE | No clear effect |
Closest papers by meaning
- Understanding and Mitigating Miscalibration in Prompt Tuning for Vision-Language Models (ICML 2025 · similarity 0.89)
- Correlative and Discriminative Label Grouping for Multi-Label Visual Prompt Tuning (CVPR 2025 · similarity 0.88)
- Revisiting Interpolation for Noisy Label Correction (AAAI 2025 · similarity 0.88)
- Towards Calibrated Multi-label Deep Neural Networks (CVPR 2024 · similarity 0.87)
- Vision-Language Models are Strong Noisy Label Detectors (NeurIPS 2024 · similarity 0.87)
- Noisy Label Calibration for Multi-View Classification (AAAI 2025 · 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.
