Research map / Detection and segmentation / Open-set and few-shot detection
Attribute Descent: Simulating Object-Centric Datasets on the Content Level and Beyond
Yue Yao, Liang Zheng, Xiaodong Yang, Milind Napthade, Tom Gedeon
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 |
|---|---|---|---|
| optimize attributes with attribute descent instead of random attributes | ObjectX synthetic data · VisDA target (synthetic only) | per-class accuracy: 65.0 -> 71.1 | Improved |
| optimize attributes with attribute descent instead of random attributes | PersonX synthetic data · Market (synthetic only) | Rank-1: +7.69% | Improved |
| optimize attributes with attribute descent instead of random attributes | VehicleX synthetic data · VeRi (synthetic only) | Rank-1: 43.56 -> 51.25 | Improved |
| augment real data with attribute-descent synthetic data instead of random-attribute data | ResNet-50 · VisDA target augmentation | per-class accuracy: 91.8 -> 93.6 | Improved |
| augment real data with random-attribute synthetic data | IDE · Duke | mAP: 58.93 -> 55.77 | Hurt |
| augment real data with attribute-descent synthetic data | IDE · Duke | mAP: 58.93 -> 59.10 | Improved |
| augment with attribute-descent synthetic data in 1:1 split | ResNet-50 · VisDA target (1:1) | per-class accuracy: 94.9 -> 95.5 | Improved |
| augment with random-attribute synthetic data in 1:1 split | ResNet-50 · VisDA target (1:1) | per-class accuracy: 94.9 -> 92.0 | Hurt |
| add orientation attribute optimization (ablation) | attribute descent · VehicleID synthetic only | mAP: -10.52% mAP when omitted | Improved |
| add camera pose attribute optimization (ablation) | attribute descent · VehicleID synthetic only | mAP: -3.54% mAP when omitted | Improved |
| add lighting attribute optimization (ablation) | attribute descent · VehicleID synthetic only | mAP: -1.6% mAP when omitted | Improved |
| add orientation attribute optimization (ablation) | attribute descent · classification and person re-ID | top-1 / mAP: 15.7% and 9.95% drop when omitted | Improved |
Closest papers by meaning
- Simulating Content Consistent Vehicle Datasets with Attribute Descent (ECCV 2020 · similarity 0.91)
- Synthetic Object Compositions for Scalable and Accurate Learning in Detection, Segmentation, and Grounding (CVPR 2026 · similarity 0.87)
- Beyond Objects: Contextual Synthetic Data Generation for Fine-Grained Classification (CVPR 2026 · similarity 0.86)
- Alice Benchmarks: Connecting Real World Re-Identification with the Synthetic (ICLR 2024 · similarity 0.85)
- Be More Specific: Evaluating Object-centric Realism in Synthetic Images (CVPR 2025 · similarity 0.84)
- Scaling Laws of Synthetic Images for Model Training ... for Now (CVPR 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.
