Research map / Vision-language and multimodal / Multimodal LLMs
When Pigs Fly: Contextual Reasoning in Synthetic and Natural Scenes
Philipp Bomatter, Mengmi Zhang, Dimitar Karev, Spandan Madan, Claire Tseng, Gabriel Kreiman
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 normal context versus no context | CRTNet · OCD | accuracy: 4% | Improved |
| congruent context versus minimal context (small objects) | CRTNet · Cut-and-paste, size 0.5-1 dva | accuracy: 17.4 -> 43.9 | Improved |
| incongruent context versus minimal context (small objects) | CRTNet · Cut-and-paste, size 0.5-1 dva | accuracy: 17.4 -> 10.6 | Hurt |
| incongruent context versus minimal context (large objects) | CRTNet · Cut-and-paste, size 7-9 dva | accuracy: 83.0 -> 66.4 | Hurt |
| use CRTNet context reasoning instead of DenseNet target crop only | CRTNet · Cut-and-paste, normal context, small objects | accuracy: outperforms DenseNet by 30% | Improved |
| use CRTNet instead of DenseNet | CRTNet · OCD | correlation with human performance: 0.66 -> 0.89 | Improved |
| add separate (non-shared) encoders for target and context (ablation) | CRTNet · OCD | accuracy | Improved |
| add confidence-weighted prediction versus context-only prediction (ablation) | CRTNet · OCD | accuracy | Improved |
| add context reasoning versus target-only prediction (ablation) | CRTNet · OCD | accuracy | Mixed |
| add gradient detachment for target encoder (ablation) | CRTNet · OCD | accuracy | Improved |
| enable self-attention in transformer decoder (ablation) | CRTNet · OCD | accuracy | No clear effect |
Closest papers by meaning
- ConTextual: Evaluating Context-Sensitive Text-Rich Visual Reasoning in Large Multimodal Models (ICML 2024 · similarity 0.88)
- ContextShift: A Controlled Benchmark for Context Dependence in Object Detection (NeurIPS 2026 · similarity 0.87)
- Evaluating Multiview Object Consistency in Humans and Image Models (NeurIPS 2024 · similarity 0.87)
- Common Inpainted Objects In-N-Out of Context (CVPR 2026 · similarity 0.86)
- ORIC: Benchmarking Object Recognition under Contextual Incongruity in Large Vision-Language Models (CVPR 2026 · similarity 0.86)
- Learning to Evolve Scenes: Reasoning about Human Activities with Scene Graphs (NeurIPS 2026 · similarity 0.86)
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.
