Research map / Detection and segmentation / Open-set and few-shot detection
Detecting Persuasive Atypicality by Modeling Contextual Compatibility
Meiqi Guo, Rebecca Hwa, Adriana Kovashka
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 |
|---|---|---|---|
| use Relative-Spatial attention instead of standard attention | unsupervised Transformer · Ads atypicality (micro avg) | AUC (%): 62.86 -> 64.32 | Improved |
| use Relative-Spatial attention instead of standard attention | supervised Transformer · Ads atypicality (micro avg) | AUC (%): 65.58 -> 66.75 | Improved |
| use Relative-Spatial attention instead of standard attention | unsupervised Transformer · Ads category OMP | AUC (%): 42.52 -> 55.18 | Hurt |
| use Relative-Spatial attention instead of standard attention | unsupervised Transformer · Ads category OIO | AUC (%): 63.07 -> 67.31 | Improved |
| use Relative-Spatial attention instead of standard attention | unsupervised Transformer · Single-Object dataset | AUC (%): 0.85 -> 0.78 | Hurt |
| increase encoder depth from 1 to 4 layers (ablation) | Transformer · Ads atypicality (micro avg) | AUC (%): 62.86 -> 64.14 | Improved |
| increase encoder depth from 1 to 4 layers (ablation) | RS Transformer · Ads atypicality (micro avg) | AUC (%): 64.32 -> 64.39 | No clear effect |
| add IoU as relative positional feature (ablation) | RS Transformer (1 layer) · Ads atypicality (micro avg) | AUC (%): 64.32 -> 64.99 | Improved |
| use RoI visual features instead of class labels (ablation) | RS Transformer (unsup) · Ads atypicality (micro avg) | AUC (%): 58.46 -> 64.32 | Improved |
| use RoI visual features instead of class labels (ablation) | Transformer (unsup) · Ads atypicality (micro avg) | AUC (%): 57.63 -> 62.86 | Improved |
| use supervised labels instead of masked reconstruction | RS Transformer · Ads atypicality (micro avg) | AUC (%): 64.32 -> 66.75 | No clear effect |
| use supervised labels instead of masked reconstruction | Transformer · Single-Object dataset | AUC (%): 0.85 -> 0.98 | Improved |
Closest papers by meaning
- ConTextual: Evaluating Context-Sensitive Text-Rich Visual Reasoning in Large Multimodal Models (ICML 2024 · similarity 0.83)
- Uncertain Multimodal Intention and Emotion Understanding in the Wild (CVPR 2025 · similarity 0.82)
- Customizing Visual Emotion Evaluation for MLLMs: An Open-vocabulary, Multifaceted, and Scalable Approach (ICLR 2026 · similarity 0.82)
- Towards Fine-Grained Interpretability: Counterfactual Explanations for Misclassification with Saliency Partition (CVPR 2025 · similarity 0.82)
- SANEval: Open-Vocabulary Compositional Benchmarks with Failure-mode Diagnosis (NeurIPS 2026 · similarity 0.81)
- A3: Towards Advertising Aesthetic Assessment (CVPR 2026 · similarity 0.81)
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.
