Research map / Detection and segmentation / Open-set and few-shot detection
Robust Subjective Visual Property Prediction from Crowdsourced Pairwise Labels
Yanwei Fu, Timothy M. Hospedales, Tao Xiang, Jiechao Xiong, Shaogang Gong, Yizhou Wang, Yuan Yao
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 |
|---|---|---|---|
| detect outliers globally with URLR instead of majority voting plus rankSVM | subjective visual property ranking · Image interestingness | Kendall tau distance | Improved |
| jointly learn outlier detection and ranking with features (URLR) instead of featureless Huber-LASSO (ablation) | URLR · Image interestingness | Kendall tau distance | Improved |
| prune outliers (URLR) instead of using all annotations (Raw) (ablation) | URLR · Image interestingness | Kendall tau distance | Improved |
| use majority voting for outlier pruning instead of using all annotations | learning to rank · Image interestingness (sparse annotation) | Kendall tau distance | No clear effect |
| increase pruning rate beyond 55% (ablation) | URLR · Image interestingness | Kendall tau distance: over 55% | Hurt |
| increase pruning rate from low values toward 55% (ablation) | URLR · Image interestingness | Kendall tau distance | Improved |
| use URLR instead of Huber-LASSO-FL on densely annotated video data | video interestingness · Video interestingness | Kendall tau distance | No clear effect |
| use URLR with added synthetic outliers (Orig+synth) | relative attribute prediction · PubFig and Scene classification | classification accuracy | Improved |
| use URLR instead of Huber-LASSO-FL at 25% error ratio | age rank prediction · FG-NET (Unint.+Int.) | rank correlation: 301 -> 546 dimensions | Improved |
| use URLR instead of Huber-LASSO-FL at 10% error ratio | age rank prediction · FG-NET (Unint.) | rank correlation | No clear effect |
| use majority voting with 5 comparisons per pair instead of Raw | age rank prediction · FG-NET (Unint.+Int.) | rank correlation | Improved |
| use URLR instead of Maj-Vot-1 after pruning rate passes 10% | age rank prediction · FG-NET (5 comparisons per pair) | rank correlation: pruning rate over 10% | Improved |
Closest papers by meaning
- Deep Robust Subjective Visual Property Prediction in Crowdsourcing (CVPR 2019 · similarity 0.93)
- Evaluating Visual Properties via Robust HodgeRank (IJCV 2021 · similarity 0.91)
- Not All Samples are Trustworthy: Towards Deep Robust SVP Prediction (TPAMI 2020 · similarity 0.87)
- Doomed to Re-Annotate, Forever: The ImageNet Story (NeurIPS 2026 · similarity 0.84)
- On the rankability of visual embeddings (NeurIPS 2025 · similarity 0.84)
- Automated Detection of Visual Attribute Reliance with a Self-Reflective Agent (NeurIPS 2025 · similarity 0.83)
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.
