Research map / Detection and segmentation / Open-set and few-shot detection
Attributes as Operators: Factorizing Unseen Attribute-Object Compositions
Tushar Nagarajan, Kristen Grauman
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 auxiliary classifier loss (ablation) | base attribute-operator model · MIT-States | harmonic mean accuracy (%): 3.7 -> 9.9 | Improved |
| add auxiliary classifier loss (ablation) | base attribute-operator model · UT-Zappos | harmonic mean accuracy (%): 20.4 -> 29.5 | Improved |
| add inverse consistency regularizer (ablation) | base model · MIT-States | harmonic mean accuracy (%): 3.7 -> 4.5 | Improved |
| add inverse consistency regularizer (ablation) | base model · UT-Zappos | harmonic mean accuracy (%): 20.4 -> 21.7 | Improved |
| add inverse consistency regularizer (ablation) | model with auxiliary loss · UT-Zappos | harmonic mean accuracy (%): 29.5 -> 29.2 | No clear effect |
| add commutative regularizer (ablation) | model with auxiliary loss · MIT-States | harmonic mean accuracy (%): 9.9 -> 11.1 | Improved |
| add commutative regularizer (ablation) | model with auxiliary loss · UT-Zappos | harmonic mean accuracy (%): 29.5 -> 33.4 | Improved |
| add antonym consistency regularizer (ablation) | model with auxiliary loss · MIT-States | harmonic mean accuracy (%): 9.9 -> 8.8 | Hurt |
| add commutative regularizer (ablation) | model with auxiliary and inverse losses · UT-Zappos | harmonic mean accuracy (%): 29.2 -> 27.5 | Hurt |
| add oracle object labels at test time | Attributes as operators · MIT-States | open world accuracy (%): 11.4 -> 49.3 | Improved |
| add auxiliary loss and trainable inputs to baseline (ablation) | LabelEmbed+ · MIT-States | harmonic mean accuracy (%): 8.2 -> 7.7 | Mixed |
| increase LabelEmbed+ layers from 1 to 3 (ablation) | LabelEmbed+ · MIT-States | harmonic mean accuracy (%): 8.3 -> 7.7 | No clear effect |
Closest papers by meaning
- Recognizing Unseen Attribute-Object Pair with Generative Model (AAAI 2019 · similarity 0.89)
- Disentangling Visual Embeddings for Attributes and Objects (CVPR 2022 · similarity 0.88)
- Improving Closed and Open-Vocabulary Attribute Prediction Using Transformers (ECCV 2022 · similarity 0.88)
- Beyond Seen Primitive Concepts and Attribute-Object Compositional Learning (CVPR 2024 · similarity 0.87)
- Compositional Caching for Training-free Open-vocabulary Attribute Detection (CVPR 2025 · similarity 0.87)
- Automated Detection of Visual Attribute Reliance with a Self-Reflective Agent (NeurIPS 2025 · 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.
