Research map / Detection and segmentation / Open-set and few-shot detection
Beyond Frequency: Scoring-Driven Debiasing for Object Detection via Blueprint-Prompted Image Synthesis
Xinhao Cai, Liulei Li, Gensheng Pei, Tao Chen, Jinshan Pan, Yazhou Yao, Wenguan Wang
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 visual blueprint instead of textual layout condition (ablation) | debiasing pipeline · MS COCO debiasing | mAP: 37.0 -> 38.9 | Improved |
| add duality-aware generative alignment (ablation) | debiasing pipeline · MS COCO debiasing | mAP: 38.9 -> 39.1 | Improved |
| add RS-driven layout recalibration (ablation) | debiasing pipeline · MS COCO debiasing | mAP: 39.1 -> 39.9 | Improved |
| add error-based dynamic debiasing (ablation) | debiasing pipeline · MS COCO debiasing | mAP: 39.9 -> 40.3 | Improved |
| use EMA momentum 0.99 instead of 0 for RS update (ablation) | dynamic debiasing · MS COCO debiasing | mAP: 38.6 -> 40.3 | Improved |
| add dynamic RS update (mu 0.99) versus static RS (mu 1) (ablation) | dynamic debiasing · MS COCO debiasing | mAP: 39.8 -> 40.3 | Improved |
| replace text-serialized layout with pixel canvas blueprint (ablation) | L2I generator · MS COCO fidelity | FID: 28.14 -> 20.15 | Improved |
| add instance discrimination to blueprint (ablation) | L2I generator · MS COCO fidelity | FID: 20.15 -> 17.05 | Improved |
| add overlap-aware rendering to blueprint (ablation) | L2I generator · MS COCO fidelity | FID: 17.05 -> 15.24 | Improved |
| use full RS (freq + visual + context diversity) instead of frequency only (ablation) | layout generation · MS COCO debiasing | mAP: 39.3 -> 39.9 | Improved |
| recalibrate position, size and category jointly instead of bias-agnostic (ablation) | layout recalibration · MS COCO debiasing | mAP: 39.1 -> 39.9 | Improved |
| add generative alignment loss for training stability (ablation) | detector training · MS COCO debiasing (3 runs) | std dev of mAP: 0.32 -> 0.06 | Improved |
Closest papers by meaning
- Multi-Perspective Data Augmentation for Few-shot Object Detection (ICLR 2025 · similarity 0.85)
- BIAS-ID: A Framework for Analyzing Transformation Biases in AI-Generated Image Detectors (NeurIPS 2026 · similarity 0.85)
- Manifold Induced Biases for Zero-shot and Few-shot Detection of Generated Images (ICLR 2025 · similarity 0.85)
- Long-tailed Object Detection Pretraining: Dynamic Rebalancing Contrastive Learning with Dual Reconstruction (NeurIPS 2024 · similarity 0.84)
- BOOTPLACE: Bootstrapped Object Placement with Detection Transformers (CVPR 2025 · similarity 0.84)
- Generative Region-Language Pretraining for Open-Ended Object Detection (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.
