Research map / Vision-language and multimodal / Human and embodied modeling
Background Invariance Testing According to Semantic Proximity
Zukang Liao, Min Chen
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 ontology-expanded keywords instead of random sampling | background sampling · Background invariance testing (250 IN9 models, N=32) | F1 score: 0.491 -> 0.650 | Improved |
| use ontology keywords instead of CLIP keywords | keyword-based sampling · Background invariance testing (250 IN9 models, N=32) | F1 score: 0.615 -> 0.650 | Improved |
| use ontology keywords instead of interval sampling | background sampling · Background invariance testing (250 IN9 models, N=32) | F1 score: 0.591 -> 0.650 | Improved |
| use ontology instead of nearest top-K sampling | background sampling · Background invariance testing (250 IN9 models, N=32) | neuron coverage (recall): 0.133 -> 0.652 | Improved |
| use ontology instead of nearest top-K sampling | background sampling · Background invariance testing (250 IN9 models, N=32) | Fleiss reliability (precision): 0.906 -> 0.649 | Hurt |
| use ontology instead of random sampling | background sampling · Background invariance testing (250 IN9 models, N=32) | Fleiss reliability (precision): 0.384 -> 0.649 | Improved |
| use ontology instead of random sampling | background sampling · Background invariance testing (250 IN9 models, N=32) | neuron coverage (recall): 0.681 -> 0.652 | Hurt |
| increase number of selected test images from 32 to 100 (ablation) | ontology sampling · Background invariance testing (250 IN9 models, N=32) | F1 score: 0.650 -> 0.652 | No clear effect |
| increase number of selected test images from 32 to 50 (ablation) | ontology sampling · Background invariance testing (250 IN9 models, N=32) | F1 score: 0.650 -> 0.651 | No clear effect |
| use Random Forest instead of AdaBoost as assessor | automated assessor · Automated assessor on IN9 model repository (RBF size 32, sigma 10) | automation accuracy (%): 74.8 -> 79.7 | Improved |
| change RBF interpolation settings from size 32 sigma 10 to size 16 sigma 5 (ablation) | Random Forest assessor · Automated assessor on IN9 model repository (RBF size 32, sigma 10) | automation accuracy (%): 79.7 -> 78.7 | No clear effect |
| use larger sigma in RBF interpolation (sigma 10 vs 2) (ablation) | AdaBoost assessor · Automated assessor on IN9 model repository (RBF size 32, sigma 10) | automation accuracy (%): 54.5 -> 74.8 | Improved |
Closest papers by meaning
- Evaluating Multiview Object Consistency in Humans and Image Models (NeurIPS 2024 · similarity 0.85)
- TTS-VAR: A Test-Time Scaling Framework for Visual Auto-Regressive Generation (NeurIPS 2025 · similarity 0.85)
- ConceptScope: Characterizing Dataset Bias via Disentangled Visual Concepts (NeurIPS 2025 · similarity 0.85)
- SANEval: Open-Vocabulary Compositional Benchmarks with Failure-mode Diagnosis (NeurIPS 2026 · similarity 0.84)
- Interpreting Object-level Foundation Models via Visual Precision Search (CVPR 2025 · similarity 0.84)
- ConTextual: Evaluating Context-Sensitive Text-Rich Visual Reasoning in Large Multimodal Models (ICML 2024 · similarity 0.84)
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.
