Research map / Detection and segmentation / CNN features
Online Data Curation for Object Detection via Marginal Contributions to Dataset-level Average Precision
Zitang Sun, Masakazu Yoshimura, Junji Otsuka, Atsushi Irie, Takeshi Ohashi
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 DetGain online sampling with strong augmentation | Faster R-CNN R50 · COCO val2017 | AP: 37.5 -> 40.0 | Improved |
| add DetGain online sampling with strong augmentation | Deformable DETR R50 · COCO val2017 | AP: 46.6 -> 48.9 | Improved |
| strong data augmentation alone (ablation) | ATSS R50 · COCO val2017 | AP: 39.2 -> 38.6 | Hurt |
| strong data augmentation alone (ablation) | Faster R-CNN R50 · COCO val2017 | AP: 37.4 -> 37.5 | No clear effect |
| add strong augmentation to DetGain sampling (ablation) | Faster R-CNN R50 · COCO val2017 | AP: 37.3 -> 39.9 | Improved |
| DetGain sampling without augmentation (ablation) | Faster R-CNN R50 · COCO val2017 | AP: 37.4 -> 37.3 | No clear effect |
| use larger teacher backbone (Res50 to Res152) (ablation) | DetGain Faster R-CNN · COCO val2017 | AP: 39.6 -> 40.0 | Improved |
| DetGain scoring instead of loss-learnability (ablation) | Faster R-CNN R50 · COCO val2017 | AP: 38.9 -> 40.0 | Improved |
| DetGain scoring instead of loss-learnability (ablation) | FCOS R50 · COCO val2017 | AP: 38.1 -> 40.9 | Improved |
| loss-based hard mining instead of uniform sampling (ablation) | FCOS R50 · COCO val2017 | AP: 38.2 -> 34.5 | Hurt |
| combine DetGain with CrossKD (ablation) | FCOS R50 (2x) · COCO val2017 | AP: 41.1 -> 42.2 | Improved |
| use uniform Beta(1,1) prior instead of fitted score distributions (ablation) | DetGain · simulated super-batches | Spearman rho: mean 0.94 | No clear effect |
Closest papers by meaning
- Plain-Det: A Plain Multi-Dataset Object Detector (ECCV 2024 · similarity 0.87)
- Simplifying Source-Free Domain Adaptation for Object Detection: Effective Self-Training Strategies and Performance Insights (ECCV 2024 · similarity 0.86)
- DetCo: Unsupervised Contrastive Learning for Object Detection (ICCV 2021 · similarity 0.86)
- SELECT: A Large-Scale Benchmark of Data Curation Strategies for Image Classification (NeurIPS 2024 · similarity 0.86)
- DetKDS: Knowledge Distillation Search for Object Detectors (ICML 2024 · similarity 0.85)
- Uni$^2$Det: Unified and Universal Framework for Prompt-Guided Multi-dataset 3D Detection (ICLR 2025 · similarity 0.85)
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.
