Research map / Vision-language and multimodal / CLIP and zero-shot
Mixture-of-Scores: Robust Image-Text Data Valuation via Three Lines of Code
Sitong Wu, Haoru Tan, Yukang Chen, Shaofeng Zhang, Jingyao Li, Bei Yu, Xiaojuan Qi, Jiaya Jia
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 data-adaptive weighting instead of plain averaging (ablation) | MoS · CC3M data filtering | Flickr30K I2T R@1: 22.7 (-1.6) | Improved |
| add uncertainty-aware temperature instead of fixed tau=1 (ablation) | MoS · CC3M data filtering | Flickr30K I2T R@1: 23.5 (-0.8) | Improved |
| use L1 instead of L2 distance for density (ablation) | MoS · CC3M data filtering | Flickr30K I2T R@1: 23.2 (-1.1) | Improved |
| use tau range 0.5-1.5 instead of 0.5-1.0 (ablation) | MoS · CC3M data filtering | Flickr30K I2T R@1: 23.9 (-0.4) | Improved |
| use tau range 0.5-1.5 instead of 1.0-1.5 (ablation) | MoS · CC3M data filtering | Flickr30K I2T R@1: 23.6 (-0.7) | Improved |
| use tau_max 1.5 instead of 2.0 (ablation) | MoS · CC3M data filtering | Flickr30K I2T R@1: 24.2 (-0.1) | No clear effect |
| ensemble only top-4 performing scores instead of all 18 (ablation) | MoS · CC3M data filtering | ImageNet-1K Acc@1: 15.6 -> 17.3 | Improved |
| ensemble bottom-4 performing scores instead of all 18 (ablation) | MoS · CC3M data filtering | ImageNet-1K Acc@1: 15.6 -> 14.8 | Hurt |
| MoS ensemble vs best single baseline score (filtering) | MoS · CC3M data filtering | ImageNet-1K Acc@1: 14.3 -> 15.6 | Improved |
| MoS ensemble vs best single baseline score (re-captioning) | MoS · CC3M image re-captioning | Flickr30K I2T R@1: 40.0 -> 43.2 | Improved |
| MoS ensemble vs best single baseline score (sample weighting) | MoS · CC3M sample weighting | Flickr30K I2T R@1: 23.9 -> 24.7 | Improved |
Closest papers by meaning
- MASS: Overcoming Language Bias in Image-Text Matching (AAAI 2025 · similarity 0.87)
- Teaching Large Language Models to Regress Accurate Image Quality Scores Using Score Distribution (CVPR 2025 · similarity 0.86)
- Calibrating Multi-modal Representations: A Pursuit of Group Robustness without Annotations (CVPR 2024 · similarity 0.86)
- Image2Struct: Benchmarking Structure Extraction for Vision-Language Models (NeurIPS 2024 · similarity 0.86)
- UMFC: Unsupervised Multi-Domain Feature Calibration for Vision-Language Models (NeurIPS 2024 · similarity 0.85)
- Q-Eval-100K: Evaluating Visual Quality and Alignment Level for Text-to-Vision Content (CVPR 2025 · 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.
